JSCC wants to fold six separately calculated futures default funds into one pool by summer 2027. ISDA’s 6 August 2026 letter names what that creates: a pro rata draw that can reach a JGB futures member’s money for a commodity default it never traded. Over 95% of the combined fund is already Index and JGB futures money.

I think this is one of the cleanest live tests available of a question every clearing member eventually has to answer: when a central counterparty pools resources across products to get more efficient, who is actually subsidizing whom, and does the subsidizer get paid for it. You do not need a JSCC relationship to use this. It travels. The same question applies at CME, Eurex, LCH, or any clearinghouse weighing the same trade, and JSCC’s own paper is the clearest worked example of it currently on the table anywhere. One template. Every CCP.

The consensus, stated at its strongest

JSCC’s own July 2026 proposal states its purpose plainly: the six funds are separated today “due to historical background of the reorganization of Japanese domestic market functions,” and JSCC wants to “further achieve both stability and efficiency” by integrating them “in light of the recent market demands for greater capital and funding efficiency.” That’s a narrower, more honest rationale than “matching global peers.” Domestic history. Not a peer claim.

A larger, pooled fund needs less collateral in total to cover the same tail risk, because losses in uncorrelated product lines partially offset each other under stress. ISDA’s own letter says its members broadly support that objective. It’s real. A member holding contributions to six separate pools is capitalized against six independent worst case draws instead of one diversified worst case, and ISDA’s letter records that JSCC’s own backtesting shows the reduction is not confined to the aggregate: every one of the six qualifications sees its own required contribution fall under the proposed framework, commodity segments included. Every segment. Lower bill. That’s the strongest fact in JSCC’s favor, and the article I drafted first left it out.

I want to give this its strongest form before I depart from it. If commodity futures and JGB futures rarely stress at the same time, for the same reason, a shared pool really can be smaller than the sum of six separate ones and still cover the same loss at the same confidence level, and every contributor pays less for the same protection. That is the entire economic case for mutualization. Not a theory. JSCC has already disclosed the number.

The variant view, and the letter that forces it

Here is what the “everyone wins” framing leaves out. ISDA is not disputing that contributions fall. It is asking whether the ones that fall the most are the ones best placed to absorb what they are newly exposed to. Cheaper is not the same question as safer.

ISDA’s letter names the mechanism in specific, mechanical terms, and JSCC’s own rulebook confirms the same structure independently. When a member defaults, other members’ contributions for that same qualification get used first, what ISDA calls the juniorisation layer. The problem sits one step further down. Once that layer is gone, the mechanism draws on every surviving member’s remaining contribution across all six qualifications, pro rata, regardless of which segment that member actually trades. A firm active only in Index Futures and JGB Futures, the two qualifications that ISDA says already make up more than 95% of the combined fund, could be called on to help absorb a Precious Metal or Petroleum Futures default it has no exposure to. Zero exposure. Still on the hook.

The reverse holds too, and it carries its own separate stake: a small commodities only member, one that trades nothing but rubber or agricultural futures, now sits behind a pro rata claim sized overwhelmingly by the financial futures book, a book it has never touched and cannot see the risk in. That member is not just exposed to a bigger pool. It is exposed to a pool whose SIZE, and therefore whose CLAIM on that member’s own capital, is set almost entirely by a business it does not do. Someone else’s book sets the bill.

ISDA isn’t asking JSCC to drop the project. Its members are broadly supportive of the efficiency goal, and the letter notes a pooled fund can also cut how often JSCC needs to call for emergency assessments from surviving members. What it asks for is a participant overlap analysis, weighted by trading volume and risk exposure, before the merger proceeds, and it recommends that if the overlap between commodity members and financial futures members turns out to be limited, JSCC should keep two funds rather than one. Two funds if the members do not overlap. One if they do.

JSCC’s own rulebook shows why the ask matters. Right now, under Article 2 of JSCC’s Rules on Required Amount of Clearing Fund, JSCC calculates six separate required amounts, one each for JGB Futures, Index Futures, Precious Metal Futures, Rubber Futures, Agricultural Futures and Petroleum Futures qualification, every business day, off a shared futures and options base date. A firm that only clears JGB futures currently owes nothing toward a rubber futures default. Nothing. Its contribution sits in its own pool, sized off its own inputs. Consolidation is not an accounting relabel: JSCC’s own July proposal states the total requirement will be “calculated as a single unit across all FIEA Clearing Qualifications”: one pool where six sit today. It changes what that firm’s posted capital is actually on the hook for.

My read: the two sides don’t actually disagree about whether the fund gets smaller, or about whether every segment benefits from that. They disagree about whether “everyone’s bill goes down” is a full answer to “who bears a tail they didn’t choose to hold,” and JSCC’s own paper, which I read directly instead of relying on ISDA’s summary, argues the first question well and does not address the second at all.

Below: the pro rata mechanism worked line by line against JSCC’s own rulebook, the LME 2022 waterfall mapped onto a JSCC-scale shock, what CME and Eurex actually do internally, the backtesting shortfall behind the precious metals margin change, and three questions to put to a JSCC relationship contact this week.

Below the paid line:

• The full pro rata mechanism worked line by line from JSCC’s own rulebook, including the yen deduction JSCC applies before sizing each member’s contribution

• The LME Clear waterfall from March 2022, mapped onto what a merged JSCC fund would face in an equivalent scenario, and the one number JSCC has already disclosed that argues the other way

• What CME and Eurex actually do internally, and why the comparison is weaker than a first read suggests

• The backtesting shortfall behind JSCC’s precious metals margin change, and the half of it most write ups leave out

• Three questions worth asking a JSCC relationship contact regardless of the comment window already being shut

Read more

Citadel and Schonfeld Strategic Advisors are reportedly running a trade that, on the ECB’s own telling, sounds like a bet on nothing. Bloomberg reported on 10 September 2026, citing people familiar with the matter, that both firms are active in it. In April, five ECB staff economists published a blog post walking through the exact phenomenon this trade prices, and their conclusion was that rising money market spreads “do not necessarily signal increasing scarcity of reserves” and “may not provide a reliable signal about the overall liquidity environment.” That’s a central bank publicly warning against misreading the thing two large funds are reportedly positioned on.

I want to take the ECB’s case seriously before I depart from it. It’s a good argument, and it isn’t the one I expected to find when I opened this research.

The ECB says the transition is orderly

The ECB’s case is specific and it’s recent. Central bank reserves have “almost halved from a peak of €4.9 trillion in 2022 to €2.6 trillion in early 2026,” and the euro area financial system “has so far adjusted well to declining reserves.” There are, in the ECB’s own words, no signs of fragmentation. No cracks yet. Short term rates, secured and unsecured both, sit close to the deposit facility rate. The spread between the euro short term rate, €STR, and the deposit facility rate has narrowed this year: 6.8bp below the DFR in the most recent reserve maintenance periods, against 7.0bp in the two before that, per the ECB’s own Economic Bulletin. I read that as the strongest single data point the ECB has for its own case.

One data point does look odd, and the ECB names it itself: 40% of overnight repo trades now execute above the deposit facility rate. Their read is that this doesn’t reflect bank funding pressure at all. Banks with thinner reserve cushions are still borrowing below the DFR on average. The trades above it are mostly hedge funds paying up for cash to fund positions elsewhere.

That’s the strongest form of the consensus I can build. Reserves are still abundant. The transition is orderly. A clean story. A trader reading a widening spread as scarcity is making a category error the ECB addressed in print, in advance, five weeks before this trade made the news. I have to grade that as a real point against my own thesis.

Two mechanisms, and the market is pricing the one the ECB didn’t write about

Here’s the observation that forces a different reading. The ECB’s blog describes the steady state relationship between the aggregate reserve level and money market rates. It never once mentions the calendar. And the calendar is exactly where the Euribor to €STR futures curve prices its widening: December, not June, not any ordinary month between them.

That gap, between what the ECB’s model covers and what the futures curve prices, is a disagreement about mechanism: which force actually produces a euro money market spike. My read is that the ECB’s April post answers a question about the LEVEL of reserves, and the futures curve is pricing a question about the CALENDAR.

An independent industry body that works alongside the Eurosystem’s own market groups names the calendar driver directly. The International Capital Market Association’s European Repo and Collateral Council put it plainly in its January 2025 year end review: the euro year end effect is driven mainly by bank balance sheet scarcity. G-SIB scores. Leverage ratio limits. Capital and liquidity requirements. Stress testing. Bank levies in some jurisdictions. All of it pushes banks to shrink their balance sheets specifically at the calendar year end, and repo books are usually the first thing cut. An easy target.

A regulatory reporting effect, dated to one balance sheet snapshot, sitting on top of whatever the aggregate reserve level happens to be that year. It can fire when reserves are comfortable. It can stay quiet when reserves are tight. Different triggers. The two mechanisms move together most of the time, which is exactly why they get treated as one. They aren’t. A trade priced on the December contract is a trade on the second mechanism, the one the ECB’s April post never addressed.

How €4.7 trillion in excess liquidity closed this basis

One term note before the arithmetic: the ECB uses “reserves” and “excess liquidity” as two related but distinct measures. Reserves ran €4.9 trillion at the 2022 peak and €2.6 trillion in early 2026; excess liquidity, the narrower figure this piece tracks most closely because it is what the futures market actually prices against, peaked at €4,748 billion and sits at €2,358 billion as of May. The gap between the two is roughly the €172 billion euro area banks must hold as minimum reserves. I use excess liquidity throughout unless I say otherwise.

Start with what Euribor and €STR actually measure, because the spread between them is the whole trade. €STR is overnight: the volume weighted average cost of unsecured overnight borrowing among euro area banks, published daily by the ECB itself. Three month Euribor is a term rate, a panel bank estimate of where unsecured interbank lending clears three months forward. In principle three month Euribor should track the market’s expected path of €STR over that window, plus a small premium for locking money up for ninety days instead of one day.

For most of the period since 2022, that premium sat close to zero. Banks were flooded with reserves. Nobody needed to bid aggressively for three month money when overnight funding was already abundant and cheap. That near zero premium is exactly what quantitative tightening is now unwinding. As the ECB lets its bond holdings under the APP and PEPP mature without reinvestment, excess liquidity has come down €2,390 billion from the 2022 peak, and the cushion that kept term funding cheap relative to overnight funding gets thinner every quarter.

The ECB’s own bank treasurer survey shows this happening bank by bank, not just in the aggregate. Banks representing 26% of euro area banking assets sat close to their internally preferred reserve level as of the Q4 2025 reading, up from 15% a year earlier, and the ECB projects that share hitting 50% by the end of 2026, a pace that has to accelerate from the trailing 11 points a year to clear 24 more points in roughly twelve months. A bank near its own target starts caring where it sources three month money versus overnight money in a way it didn’t two years ago, because term funding also carries regulatory value: the liquidity coverage ratio and the net stable funding ratio both reward longer tenors, and the survey shows banks defending internal targets set above the regulatory minimum. Their own cushion, not the rulebook’s.

Add the calendar and a second mechanism switches on

LCR, NSFR, the leverage ratio, G-SIB systemic scores: every one of those gets measured against a balance sheet snapshot at 31 December. ICMA calls 2016 the worst EUR REPO turn on its own ten year record, a “perfect storm” of bond market positioning, a shortage of high quality collateral and onshore/offshore dollar imbalances that practically closed the market. That is a claim about the secured repo market specifically, one that runs on collateral, and it predates most of the regulatory ratios that now drive the unsecured effect this piece is about.

The unsecured precedent runs older still. I checked that the 2007 turn was serious enough that the ECB itself intervened directly in the interbank market, reinforcing its allotment policy and stretching a mid December refinancing operation to a two week maturity specifically to bridge Christmas and the turn in one operation. I could not independently verify a precise basis point figure for that jump against a primary source I trust, so I am not going to quote one. The ECB’s own emergency response is evidence enough that the mechanism is real and older than the rules that now formalize it.

A three month Euribor contract settling in December prices the risk that banks pull back from term lending exactly when regulatory reporting makes balance sheet space most expensive. A €STR future, an overnight rate, carries none of that seasonal weight. Nothing to bridge. The spread between them is where the secular decline in reserves and the cyclical balance sheet squeeze actually meet, and it’s the only instrument I know of that isolates the second effect from the first.

This isn’t a new trade either. MNI Markets, an independent fixed income wire with no connection to Bloomberg’s September reporting, recorded elevated volume in this exact spread on 31 July 2024, trading at 10.75 basis points against the March 2025 Euribor contract, on volume its own desk described as running two to four times what the other contract months traded that same session. Desks were running this position more than a year before it made the news. Ahead of the story, not behind it.

Below the paid line:
- The full ECB rate and liquidity table (€2,353 billion deposit facility recourse, €172 billion minimum reserves, €24 billion in MRO/LTRO usage) worked against the reserve scarcity threshold the bank treasurer survey implies
- The exact ICE contract construction a euro rates desk uses to hold this spread into December, sized against its 106,910 lot open interest record and its 90%-plus margin offset versus Euribor
- The 2024 AND 2025 near misses worked in full, with the specific ICMA sourced reasons the priced spike deflated both times, and the secured versus unsecured distinction the thesis actually rests on
- The reflexivity problem, worked through against the 40% above-DFR repo share the ECB itself attributes to hedge fund cash demand
- The 31 December to 2 January settlement window and the exact basis point level that proves this trade right or wrong

Read more

Watch the video (10:54): Why Morgan Stanley’s Derivatives Book Jumped $55 Trillion. The video includes corrected bank rankings from the OCC’s Q1 2026 tables.

Four US banks, JPMorgan Chase Bank NA, Citibank NA, Bank of America NA and Goldman Sachs Bank USA, held 79.1% of the entire US banking system’s derivatives book at the end of Q1 2026, down from 87.1% a year earlier, while the total book itself exploded 42.5% to $296.5 trillion.

Below the paid line:

• The full five quarter concentration and notional series, worked against each other quarter by quarter, including the Q3-to-Q4 peak and drop that a documented capital rule seasonal pattern predicts, and the same quarter year over year check that tests whether that pattern is the whole story.

• Which of the four dealer banks carries the highest capital cost per dollar of book, worked from the Fed’s own June 2026 surcharge table against each bank’s OCC reported risk based capital, with the stress buffer and GSIB surcharge shown as separate addends.

• The single quarter value at risk move that tells you which of the four actually leaned into this quarter’s volatility, and which three sat it out.

• The Federal Register’s own quantitative estimate of how much the pending GSIB surcharge rewrite would cut capital costs at these exact four banks, and what that does to this whole story once it finalizes.

• My worked arithmetic for why a thirteenfold FX revenue jump is hard to square with the deliberate pullback story, and the one piece of documented literature about these same banks that argues the other way.

Below the gate: the Fed’s own $23 billion GSIB surcharge estimate, and the per-bank capital math behind Goldman’s $147 million VaR jump.

The consensus: capital is the moat, and it is real

Ask a derivatives desk why four banks dominate the US market and you get a clean answer. Capital. A swap dealer has to hold collateral, mark exposures daily, absorb margin calls, and carry the position on a balance sheet a regulator stress tests every year. Only a handful of banks have balance sheets large enough, funding costs low enough, and risk infrastructure sophisticated enough to do that at real scale.

Post 2008 capital rules made this explicit. The Global Systemically Important Bank surcharge adds a permanent capital tax on top of the ordinary minimum, scaled to a bank’s size, interconnectedness and complexity. Derivatives activity feeds directly into that score. JPMorgan carries a 4.5% surcharge, the highest of any US bank. Size costs the most. Citigroup and Goldman Sachs carry 3.5%. Bank of America carries 3.0% (Federal Reserve, Large Bank Capital Requirements, June 2026, Table 1). Add the three pieces and the total shows its own arithmetic: the 4.5% minimum CET1 floor, plus a stress capital buffer, plus the surcharge. JPMorgan’s stack is 4.5 + 2.5 + 4.5 = 11.5%. Citigroup’s is 4.5 + 3.6 + 3.5 = 11.6%. Goldman’s is 4.5 + 3.4 + 3.5 = 11.4%. Bank of America’s is 4.5 + 2.5 + 3.0 = 10.0%.

That is a large, durable moat, and I don’t dispute it. Regional banks and most foreign dealers cannot fund a derivatives book against those requirements and still clear a competitive return on equity. The market consolidated into a small dealer club and stayed there for over a decade. The moat is real. This is the correct baseline. What I’m arguing is that the baseline just moved, and the consensus has no mechanism ready to explain why.

It also explains why this is an American story specifically. Risk.net’s own Dealer Rankings 2026 coverage frames it that way: the top US houses expanding while European names lose ground on the same buy side books (Risk.net, “How US dealers flipped the script in OTC trading”, Dealer Rankings 2026 series). The same Fed table that sets JPMorgan’s surcharge also lists the US arms of foreign banks operating here, and the gap is not subtle. Deutsche Bank’s US intermediate holding company, DB USA Corporation, carries no GSIB surcharge at all, and still lands at a total CET1 requirement of 16.0% (4.5 + an 11.5% stress capital buffer), the single highest figure anywhere in the Fed’s table, above JPMorgan’s 11.5% and Goldman’s 11.4% (Federal Reserve, June 2026, Table 1). UBS Americas Holding LLC sits at 9.7%. HSBC North America sits at 9.6%. Home turf costs less. A European dealer running derivatives out of its US subsidiary pays a steeper toll than the American incumbents do, before its home regulator charges it anything at all. That toll alone would keep a European bank’s US book smaller than its balance sheet could otherwise support, independent of anything that happened this particular quarter. Capital, not talent, draws this line.

The full five quarter series, and the shape nobody had drawn

The OCC states the current number in one sentence: “the four banks with the most derivative activity hold 79.1 percent of all bank derivatives” (OCC Q1 2026, p.11). I pulled every quarter behind it, not just the two that made the cleanest headline. Here is the full run, concentration alongside total notional:

This is not a smooth drift, and I am not going to describe it as one. Concentration actually rose in Q2, to 87.3%, before it started falling. Notional rose for two straight quarters to a peak of $231.8 trillion in Q3, then dropped 10.3% in Q4 to $208.1 trillion, before exploding 42.5% in Q1 2026 to a level 28% above that Q3 peak. A rise. A drop. Then a spike. Not a quiet three quarter slide into a sudden cliff.

That Q4 drop followed by a Q1 rebound is not a shape I get to treat as neutral. That shape has a name. It is the exact seasonal pattern documented in the literature on how GSIBs manage their surcharge scores, and I address it directly in the next section rather than skip past it.

I expected the boring explanation here: new entrants diluting the top four’s share by simple arithmetic. It isn’t there. I checked the count quarter by quarter: 1,213 in Q1 2025, a peak of 1,221 in Q3 2025, then 1,197 and 1,192. Net change over the year is 21 fewer banks, not hundreds more. Same banks. Bigger pie. Smaller slice for the top four.

Where the $88.4 trillion actually came from

I want to know where the $88.4 trillion actually came from before I trust any story built on top of it. Interest rate notional rose $67.4 trillion in the quarter, from $135.8 trillion to $203.2 trillion, a 49.7% jump (OCC Q1 2026, p.11, Table 10). One instrument carried 76% of the entire quarter’s growth. Interest rates did the heavy lifting. FX added $18.1 trillion. Equity added $1.1 trillion. Credit added $1.6 trillion. Commodity added $0.2 trillion. Rates dwarfed everything else. By contract type, swaps carried two thirds of it: $58.3 trillion of the $88.4 trillion increase, taking swaps to 61.8% of all bank derivatives notional (Table 11, same page). One instrument, one contract type. Futures, forwards and options split most of the remaining third between them.

This is the part that should stop you. Interest rate notional exploded. Interest rate trading revenue collapsed, down 87% year over year to $1.1 billion. If a bank were putting on more directional risk and getting paid for it, revenue and notional would move together. My read is that they did not, because this was not that kind of growth. Volume without pay. That’s the tell.

I want to be precise about what the clearing data actually shows, because it is more mixed than a clean compression story. Centrally cleared interest rate notional rose from about $61.0 trillion to about $84.5 trillion this quarter, a real increase. But its SHARE of the total fell, from 44.9% to 41.6%, because uncleared bilateral notional grew even faster, from roughly $74.8 trillion to $118.7 trillion (OCC Q1 2026, p.12, Table 12). Novation compression, where one bilateral trade splits into two offsetting cleared legs, would push the cleared share UP, not down. That is not what happened. The share fell instead. So I am not going to lean on the clearing table as proof of a compression story specifically.

What the data does show is growth spread broadly across both the cleared and the uncleared book at once, not concentrated in one structure. I read that breadth, cleared and bilateral both expanding, as evidence the flow is client driven and widely distributed rather than one desk’s proprietary position. That kind of flow does not need JPMorgan or Goldman specifically. It needs a swap dealer with a balance sheet, cleared or not. More than four banks qualify for that narrower job, even if only four can run the full book.

Read more

Coinbase’s September 1 filings to relaunch perpetual futures on individual US stocks read as a demand story, but a capital rule window and a live federal lawsuit sit underneath it.

I’d size this as an open question, sitting on top of a demand story I’m not trying to argue away. I’ll show exactly where my own paper trail runs dry, and where a red team I ran against my own draft actually broke a claim.

The consensus, stated at its strongest

The trade press has the demand story right, and I want to give it its full weight before I complicate it. Coinbase’s chief policy officer, Faryar Shirzad, said it plainly: “Equity perps have proven demand internationally, and we’re excited at the prospect of a regulated pathway for U.S. investors.” That’s not spin. Combined crypto perpetual futures volume rose from $4.14 trillion in January 2024 to $7.24 trillion in January 2026, per CoinGecko Research. Perpetual swaps now carry roughly three quarters of all crypto derivatives flow. Coinbase itself isn’t new to this specific product: it launched single stock perpetuals for non-US customers in March 2026, building on the $2.9 billion Deribit acquisition that closed in August 2025. I take all of that as settled. The demand is real. Anyone claiming this is a paperwork stunt with no real user behind it is wrong, and I’m not that person.

The engineering case holds up too. A dated future converges to spot at expiry through cost of carry. That convergence is where a quarterly market bleeds liquidity: open interest fragments across contract months, and the front month thins as expiry nears. A perpetual swaps the roll for a funding rate, an interest rate plus a premium index, paid directly between longs and shorts on a fixed schedule. One instrument stays honest continuously. Four instruments used to stay honest four times a year, imperfectly. If the roll caused the friction, killing it should fix the friction. A CFA charterholder reading only the press coverage would land here and stop.

My variant view: what demand alone can’t explain

Here’s where I get off consensus. If proven demand alone explained this wave, it would explain why single stock leverage products exist in 2026. It would not explain why Coinbase’s specific notice registration landed on September 1 rather than August 15 or October 3, and the trade coverage doesn’t ask that narrower question at all.

Start with the history the demand story skips past. Single stock futures already existed in the US for nineteen years as a listed product, traded on OneChicago, an exchange CME and Cboe founded in 2001 with Interactive Brokers joining as a partner in 2006. They didn’t fail from indifference. The Commodity Futures Modernization Act of 2000 created the legal category “security futures product,” and the same statute ordered regulators to set its margin so it could never sit below the lowest margin on a comparable exchange traded option. Congress built that floor specifically to stop a future from becoming a cheap synthetic option. The SEC and CFTC implemented it in 2002 at a flat 20% of contract value.

That statute didn’t stop OneChicago from building a real market. Annual volume climbed from roughly 6.3 million contracts in 2012 to 14.9 million in 2017, its record year, across five straight years of growth. This was working order flow. Real money, real size. Then volume fell 52.6% in 2018, to 7.1 million, and kept falling every year after. I haven’t found a single documented cause for that collapse, and I want to be direct about a second explanation I found while checking my own margin story.

Derivatives writer Jakub Rehor’s account of the period puts real weight on a distribution failure specific to security futures. A broker wanting to offer them had to also register as a CFTC futures commission merchant under a separate, contradictory compliance regime, and most securities broker dealers looked at dual registration and declined. Security futures also drew ordinary income tax treatment instead of the 60/40 capital gains treatment normal futures get, a second handicap margin alone doesn’t capture.

I’m not walking back the margin story. Congress’s floor is real, dated, and it removed the one advantage that would have made the product worth the distribution hassle. Margin was one blade of a pair of scissors. Not the whole cut. Two causes, not one.

It matters because Coinbase’s structure answers the other blade more precisely than it answers margin. Form 1-N registers the exchange. Form BD-N registers the broker dealer. One corporate family holds both, which is exactly the dual registration wall that made brokers walk away from OneChicago in the first place.

Two dates that make this filing worth a second look

On October 22, 2020, at that first joint meeting, the SEC and CFTC voted to cut the margin floor to 15%. OneChicago had told members on August 13 it was closing. Its last trading day, September 18, landed five weeks before regulators voted to approve that cut, and the rule did not formally take effect until December 24, 2020, three months after OneChicago was gone. The one market that could have tested whether 15% was low enough closed before the number existed, and stayed closed for nearly six years. Six years of silence. I’ll say now, before the objection finds me later: CME relaunched a dated single stock future this July under this exact 15% floor, so the floor has been tested since. What it has not tested, not once, in nineteen years of the category’s existence, is a version with no expiry at all. That is the specific, narrower claim the rest of this piece needs, and it is the one I can actually defend.

Now the second half of the timing. On June 30, 2026, the SEC and CFTC jointly asked for comment on further implementation of portfolio margining and cross margining of securities and derivatives, an explicit continuation of the same 2020 project. SEC Chairman Paul Atkins called cross margining an opportunity to free up “liquidity that remains frozen in separate accounts.”

That project is older than the June date suggests. SEC Commissioner Hester Peirce was already commenting on this same harmonization push in a September 2025 statement, a full year before this specific window opened. The June 30 request is a discrete step inside a much longer running conversation. That window closed August 31, 2026, and Coinbase filed Form 1-N and Form BD-N the very next filing day.

I have not found a comment letter from Coinbase or any subsidiary in that docket, and I do not have Coinbase’s internal product calendar, so I’m holding this as a strong circumstantial marker, not a proven cause. The honest version: a company with an obvious, independent, demand driven reason to file in 2026 chose to file on the one business day when a related capital rules proceeding had just gone quiet for deliberation. That is worth noting. It is not proof of intent.

The mechanism: what actually changes, and what doesn’t

Start with what a perpetual actually fixes. Crypto exchanges settle funding every eight hours. Coinbase’s filings do not specify a cadence for the equity product, and that’s one of the real unknowns in the paper trail. The mechanism holds regardless of frequency, though. When the perpetual trades above spot, longs pay shorts. That pulls in fresh selling and drags price back down. Below spot, the payment flows the other way. Continuous funding does the job periodic rolling used to do, only without the calendar. That much is a real fix. Margin is a separate question entirely.

It does not touch the margin problem, and I have not seen this point made anywhere in the coverage. “Security futures product” is defined by what sits underneath the contract, a single security or a narrow based index. It says nothing about whether the contract expires. Exchange Act Section 3(a)(55) and CEA Section 1a(45) both define the category by the underlying instrument. Neither the 2002 rule nor the 2020 amendment carves out a no expiry structure. If that reading holds, a Coinbase single stock perpetual sits on identical legal footing to an OneChicago contract from 2019: same joint SEC-CFTC oversight, same 15% floor, unless the pending rulemaking moves it.

The forms Coinbase actually filed confirm which regime it walked into. Form 1-N is the specific notice registration a CFTC registered designated contract market files with the SEC to become “notice registered” under Section 6(g), a simplified path Congress built only for exchanges listing security futures products. Form BD-N is the parallel registration for a broker dealer trading exclusively in them, under Section 15(b)(11). Coinbase did not invent a fresh, unregulated crypto wrapper. It filed the two forms the 2000 era regime built for this exact job.

The entities behind the filing, and the doors they have already used

It filed them through entities with a live record in this kind of doorway. Coinbase Derivatives, LLC is the CFTC registered exchange Coinbase bought in 2022 as FairX, already running nano Bitcoin and Ethereum futures. Coinbase Financial Markets, Inc. separately holds a CFTC no action letter, granted May 29, 2026, letting it post customer crypto assets as margin at its affiliated foreign venue, Deribit FZE, for perpetual contracts the CFTC agreed to treat as foreign futures. That same May 29 letter cites a parallel CFTC order approving KalshiEX LLC’s BTCPERP futures contract, a second, independent perpetual futures approval moving through the Commission the same day, on an entirely different rail. I would read that as the whole apparatus around perpetual style products loosening at once. Not one filing. A pattern. Coinbase’s own move here is at least its second doorway in twelve months.

And here’s the part the margin story alone misses: one corporate family holding both the exchange registration and the broker dealer registration is the more precise fix for OneChicago’s actual, documented failure mode. The dual-FCM-and broker dealer wall that made securities firms decline to distribute security futures in the 2000s doesn’t exist inside Coinbase’s own structure the way it existed for an independent broker deciding whether OneChicago’s product was worth the compliance cost. That is real. It has nothing to do with margin.

Run the arithmetic on what 15% still costs you, margin fix or not. A trader who wants $10,000 of directional exposure to one stock puts down $1,500 under the security futures margin rule. The same trader buying a comparable option struck at the money typically pays somewhere in the range of 4% to 8% of notional in premium (a rough illustration using standard option pricing approximations, not a quoted market figure), depending on the stock’s implied volatility and the option’s tenor. That premium is the trader’s entire maximum loss, not a refundable deposit.

On volatility alone, options can beat the future on the same directional bet. Options still win here. And the option owner can’t be called into forced liquidation overnight the way a futures holder can. That’s the actual competition security futures have never won. Never once. Fifteen percent is progress against 20%. It’s not obvious progress against a four percent option premium on a volatile name, which is exactly the kind of single stock a retail trader wants leveraged exposure to in the first place.

One more door exists already, for accounts that can reach it. FINRA Rule 4210(g), the industry’s portfolio margin framework, names “a securities futures product” among the instruments eligible for portfolio margining, alongside listed options and unlisted equity derivatives. Portfolio margin prices a whole position’s net risk instead of a flat per contract percentage, and it can push effective margin on a hedged book well under 15%. That relief isn’t hypothetical. It exists today, for sophisticated or institutional accounts, not the retail trader Coinbase’s own marketing is aimed at. The open question I’d flag here is whether the current SEC-CFTC proceeding extends anything like portfolio margin economics down to an ordinary account. That’s the only version of this product that actually needs 24/7 access to justify itself.

I built a paid research note that goes further on exactly this question: the specific public query that tracks the margin docket in real time, plus the worked arithmetic for a portfolio margined book.

The clock this filing is actually racing

There’s a fourth date that belongs beside the other three, and it’s the one that already answers part of my own earlier claim. CME itself already re entered this market once this year: it relaunched cash settled, dated single stock futures on 55 US names, SpaceX and Nvidia among them, on July 27, 2026, five weeks before Coinbase’s filing, and CME’s own product notice sets the minimum margin at 15%, the identical floor this entire piece has been discussing.

That relaunch is exactly what Bloomberg reported CME is betting on: “the rise of retail trading and today’s market environment of hot IPOs with limited share availability,” with CME’s global head of equities, Tim McCourt, quoted separately calling it a way to “bring in a lot of new traders” to CME’s ecosystem. That’s a second, independent, on the record demand explanation I have to weigh rather than wave off. I’ll come back to what that does to my thesis in the next section.

CME’s own August volume release breaks out interest rate, equity index, energy, agricultural, metals, FX and cryptocurrency as named lines. Single stock futures don’t get one. I read the omission as informative on its own: the product exists, is trading under the exact floor this piece is built on, and isn’t yet large enough for CME to name in its headline numbers five weeks in.

CME is also, right now, suing the regulator that would have to bless Coinbase’s structure, though not over the question I first assumed. On June 18, 2026, CME filed suit against the CFTC in the DC district court, docketed as case 1:26-cv-02157 before Judge Colleen Kollar-Kotelly, arguing that the CFTC’s May 29 order classifying Kalshi’s Bitcoin perpetual as a future rather than a swap sidestepped the correct process. That fight is specifically about Bitcoin, a commodity under the CFTC’s own jurisdiction, under the Commodity Exchange Act’s swap versus future line. Both possible outcomes keep the product inside the CFTC’s rulebook.

It does not directly govern Coinbase’s equity product, which sits on a different statute entirely: the Exchange Act’s “security future” category Coinbase invoked with Form 1-N and BD-N, versus “security based swap” under Title VII of Dodd-Frank, which is SEC-exclusive jurisdiction with its own capital and clearing regime. That fork is separately, actively contested. The Hyperliquid Policy Center is lobbying regulators right now to have cash settled equity perpetuals classified as security futures rather than swept into the broader swap category, which only makes sense if that classification is still open.

The CFTC moved to dismiss CME’s suit on September 2, one day after Coinbase’s own filing, calling CME’s claim of competitive injury “much ado about nothing.” CME’s opposition is due October 2, 2026.

I’d treat the equity specific classification fork, not the Bitcoin lawsuit directly, as the real capacity constraint on this whole opportunity. The worst case for Coinbase isn’t that the margin floor stays at 15%. It’s that an equity perpetual gets pushed into the security based swap regime entirely, which the article’s margin discussion above doesn’t come close to describing and which would be a materially larger problem than anything in this piece’s numbers. The Bitcoin case still matters as a signal of which way the Commission is leaning on perpetual structures generally, and Judge Kollar-Kotelly’s eventual ruling is worth watching for that reason. It just isn’t the direct legal mechanism I originally read it as.

What separates my read from the simpler one

Here’s the strongest version of the objection, and I want to state it the way a sophisticated reader actually would: two different companies converged on the same product category in 2026 off the back of an obvious, already measured retail boom, so you don’t need a regulatory window story to explain why either of them showed up this year. Leveraged single stock ETFs grew from roughly $17 billion to somewhere in the $43 to $65 billion range in under two years.

CME’s own leadership put a name and a quote on record attributing its identically margined relaunch to retail demand and hot IPOs, not to any rulemaking. Coinbase has been running the same offshore product since March and telegraphing stock perpetuals as a roadmap item for over a year, off an acquisition that closed thirteen months before this filing.

I think that objection is right about the category. It is too narrow about the date. Proven demand explains why leveraged single stock products exist in 2026 at all. It doesn’t explain why Coinbase’s specific US notice registration landed on September 1 rather than any other business day that quarter, and the demand story has nothing to say about margin economics either way, because CME’s identically margined relaunch and Coinbase’s own filing are both still operating inside the same 15% floor whether or not a rulemaking ever moves it. OneChicago is the control case that keeps this distinction honest: 14.9 million contracts in one year is real, cleared demand, and the product still died because demand was never the binding constraint, capital cost and distribution structure were. A crypto native audience with proven appetite for perpetuals answers the demand question. It says nothing about whether the US-regulated version clears the capital hurdle that broke this exact legal product before, or survives the classification fork that’s live right now. Demand, capital economics, and legal classification are three separate conditions. This piece is about the second and third, not a rebuttal of the first.

Three things I can’t rule out

First, I haven’t located the text of Form 1-N or Form BD-N themselves. Coinbase disclosed them as images on its own account rather than through EDGAR. As of this writing, I could not confirm the phrase “equity perpetual” or “security futures” anywhere in Coinbase Global’s own EDGAR full text record, including what I could find of its most recent 8-K, filed September 2. I was not able to independently rule out a filing landing after my own search, so treat that specific negative as unconfirmed rather than settled.

Second, I haven’t found a comment letter from Coinbase or an affiliate in the SEC-CFTC harmonization docket. My “positioned to benefit” read is inference from a one day gap between two dated events, sitting inside a harmonization conversation that’s been running for at least a year. It isn’t a documented lobbying position I can point a reader to directly.

Third, and the one that matters most: no SEC or CFTC release I’ve found states outright whether a perpetual, no expiry structure gets margined identically to a dated one under the existing rule, or whether it risks being pushed into the security based swap category the Hyperliquid Policy Center is actively lobbying against. My reading follows from how the statute defines the product category by its underlying security rather than its settlement calendar. It’s never been tested by an actual ruling, because no perpetual security futures product has existed to test it on. I’m not hiding that hole. It’s the argument. The untested case is what makes this filing a bet instead of a formality.

What would change my view

The claim above is falsifiable on purpose, and three dated, public checkpoints settle it, in order of how soon each one lands. CME’s opposition brief is due October 2, 2026, and Judge Kollar-Kotelly’s eventual ruling is the nearest signal, even though it governs the Bitcoin fork directly rather than the equity one. If the harmonization rulemaking produces a final rule, or an explicit decision not to act, that leaves security futures margin at 15% with no expanded portfolio margin path for ordinary accounts, and Coinbase’s product still launches to volume clearing OneChicago’s 14.9 million contract peak within its first twelve months of listing, then demand and crypto native distribution are doing the work I attributed to margin and classification timing. I’d be wrong about the mechanism even if right about the filing pattern. Separately, the actual next procedural step after a notice registration is a CFTC product self certification. If that filing states outright how the product is margined and classified, my open question resolves by disclosure instead of by inference.

What to watch instead of the launch date

OneChicago didn’t die of low demand. It built a real, growing market across five years, then lost most of it inside twelve months, carrying both a capital cost Congress wrote into the statute on purpose and a dual registration wall that kept most brokers away entirely. The one fix regulators made to the capital side arrived after the only market for it had already closed, and stayed untested for the no expiry version of the product for six years after that. Coinbase isn’t filing into settled ground, and it isn’t filing on demand alone either. It’s filing through a structure that solves OneChicago’s distribution problem more precisely than its margin problem, in the same month a related capital rules proceeding went quiet for deliberation, while a live lawsuit works out whether perpetual futures survive as futures at all.

That’s the actual trade sitting inside this filing. The funding rate engineering is real, and it solves a real problem. The demand is real too. Whether the capital and classification questions get answered in Coinbase’s favor is a question for regulators and a federal judge, not for engineers or for how big the crypto perpetuals market already is. A harmonization rule that hands security futures a genuine capital edge over listed options, combined with a classification ruling that keeps equity perpetuals out of the security based swap bucket, gives Coinbase a path OneChicago never had. Absent either one, a perpetual single stock future is a better built, better structured version of a product that already failed twice for reasons that have nothing to do with how often it rolls.

Watch the Federal Register and the D.C. district court docket. Don’t watch Coinbase’s app store listing. That’s where this actually gets decided.

The same pattern shows up elsewhere in market structure right now. The CFTC rejected a special call reporting plan from the futures industry in 2011 and adopted the same plan as a transitional bridge fifteen years later, in 2026, which is the same story as a margin rule cut five weeks after the market it was meant to save had already closed. And the clearinghouse side runs the identical asymmetry in the other direction: Kalshi’s clearing members can be assessed up to 550% of what they put in, while the exchange’s own default fund carries no matching obligation to refill. Capital rules built for one product rarely retire on the same schedule as the product itself.

I break down filings like this one on YouTube and post shorter research notes on LinkedIn.

The Decision-Grade Version

This piece is complete on its own. The thesis, the evidence, and what would kill the view are all above, and nothing was held back to sell you a next step.

The Patreon note is a separate piece of work rather than a deeper cut of this article. It takes one tool from inside this story, i.e. the exact public query that tracks the margin docket in real time, and writes out the worked arithmetic for what a portfolio margined book actually costs at 15% versus 20%. Written for people who want to watch this resolve as it happens rather than read about it after.

→ Read the margin tracker note

→ Or join the Patreon community for every note

Cover photograph: Teemu008 from Palatine, Illinois, CC BY-SA 2.0, via Wikimedia Commons.

MSCI modeled a $310 million passive bid for Adani Energy Solutions. The 31 August closing auction needed $569 million to clear, and the stock fell 10.4% and stayed there.

Below the paid line:

• GQG’s own quarterly and SAST filings, read side by side against each other. What they can and cannot prove about who actually sold in the auction print

• Nuvama’s passive demand build reconstructed for all four MSCI adds, and what it implies about the size of money that was supposed to show up

• the auction’s own capacity math, worked from the $4.1bn figure and BNP Paribas’s disclosed share of it

• the fragility ratio worked two ways, session total against modeled bid and identified print against modeled bid, and whether either is now the going rate

• why Adani Group’s own post-Hindenburg volatility record, running since January 2023, can’t settle whether CAS or the stock’s concentration explains the size of the move

I make that 1.84 times, on the number that actually reconciles. $569m changed hands in the session against a modeled bid of $310m. The session alone was nearly twice the size of the entire passive order the trade was built to absorb, before counting who specifically sold into it.

Below the paid line: GQG’s two filings, reconciled line by line, and the 1.84x fragility ratio for the next MSCI rebalance.

The consensus: Shah and Kejriwal both called this manageable

The professional view of index rebalancing arbitrage is not naive, and I want to state it in a form its own practitioners would sign. Event driven and quant funds have run this trade since MSCI, S&P and FTSE reviews became transparent, rules based events.

Antti Petajisto’s index premium research, covering 1990 to 2005, measured the mechanism directly. S&P 500 additions carried an average 8.8% price impact. Deletions averaged minus 15.1%. Both real, both measured. A fund that buys ahead of the effective date and sells into the passive bid is not gambling on direction. It is monetizing a flow it can see coming, from a counterparty with no discretion to walk away.

Tejas Shah of Equirus Securities put the calmer version to the wires days before the rebalance. The larger names entering the MSCI basket were not major index components, he said, so volatility at the headline level would not be very significant. Called it early. Anand James at Geojit Investments said almost the same thing after the fact, in fewer words: flows driven moves were always expected.

Both are describing a market that did what it was supposed to do. A predictable amount of buying showed up for a predictable reason, on a predictable day. Nothing unusual, on the surface. The Nifty Bank index closed about 0.9% higher after dipping into the auction. That is the orderly outcome the consensus predicts.

It is a coherent, evidence backed position, and most of the time it is correct. It just wasn’t the whole story for Adani Energy Solutions.

The variant view: a session nobody modeled for

The session needed $569m. The model only ever promised $310m. Here’s the observation that forces a different reading. Nuvama modeled $310m of passive demand landing in Adani Energy Solutions from trackers alone. $569m of the stock changed hands in the session that was supposed to absorb it. Nearly double. That’s the session level mismatch. It alone is 1.84 times the modeled bid.

Inside that session, Bloomberg’s reporting names a single entity that sold 21.92 million shares at Rs 1,417 in the closing print. At that day’s official Rs 83.15 to the dollar, published by the Federal Reserve, that print alone comes to roughly $373.6m. That’s already bigger than the whole modeled bid, before counting anyone else who sold. I want to be precise rather than round up: it isn’t quite the “nearly double” a loose read of the session total might suggest. Call it a fifth larger. The rest of the 1.84x gap came from elsewhere in the session’s turnover, not from this one print.

My read: the index rebalancing arbitrage trade has a structural blind spot, and CAS just made it visible. The trade prices one side of the ledger with real rigor. Nuvama’s $310m and $202m figures for Adani Energy Solutions and Adani Enterprises aren’t guesses. They’re built from MSCI’s published float adjusted weight changes, multiplied against the assets under management of every fund that tracks the index, and a second, independent estimate lands on the same two figures. Two sources, one number.

The other side of that window gets modeled as noise. Active sellers who might show up for reasons that have nothing to do with the index review. If they get modeled at all.

GQG Partners is the unnamed entity the market read into Bloomberg’s reporting, and it had every reason to sell some Adani Energy Solutions that week for reasons that had nothing to do with MSCI. It built the position starting 2 March 2023, the first major institutional buyer across the Adani group after the Hindenburg report, putting roughly $1.9bn, Rs 15,446 crore, to work as a public vote of confidence. A big bet, publicly made. It’s been trimming that position for over a year, and it isn’t just Adani Energy Solutions. Its own quarterly disclosures show it also cut Adani Enterprises, Adani Green Energy, Adani Ports and Adani Power the same quarter. That pattern is the tell. A multi name, multi quarter drawdown reads like ordinary portfolio management, on its own calendar, that happened to land in the one twenty minute window each quarter where its size matters most.

A less charitable read is available, and I want to name it rather than let the coincidence framing do all the work. A fund unwinding size that large doesn’t usually choose to clear it inside a brand new, thin, twenty minute auction if the goal is a clean print. It’s at least as plausible the seller timed the trade to the rebalancing on purpose, using the modeled passive bid as cover for size that would otherwise be hard to clear without moving the market itself. I can’t distinguish “coincided with the passive bid” from “was aimed at the passive bid” from public filings. That’s a real limit on the ordinary portfolio management reading, and I’m naming it rather than hiding behind it.

The mechanism: how twenty minutes turned a session into a 10.4% print

Five steps turn a compressed auction window into a broken print. Walk the chain from the auction’s design to the printed loss, because “the trade was crowded” is not an explanation on its own. Crowded trades survive uneventful days all the time.

Step one: CAS compresses a day’s information into twenty minutes. NSE’s Closing Auction Session launched 3 August 2026, running 3:15pm to 3:35pm on every trading day. Instead of a stock’s official close being whatever price continuous trading last produced, buy and sell orders accumulate for that twenty minute window and the exchange computes a single equilibrium price. On 31 August that window cleared $4.1bn to $4.2bn of MSCI linked turnover. That is 21 to 22% of NSE’s entire cash market volume for the day. A fifth of the day, in twenty minutes. A mechanism built to concentrate price discovery had, on its first real stress test, concentrated an entire quarter’s index review flow into minutes.

Step two: one bank cleared nearly half of it. That is a depth problem stated as a fact, though not a complete one. BNP Paribas alone executed about Rs 180bn, roughly $1.9bn, of the day’s $4.2bn CAS total. Nearly half the entire auction’s flow ran through one counterparty’s book. The reporting doesn’t say whether that was BNP risking its own balance sheet or routing a long list of institutional clients through its execution desk. Two very different stories. The two mean different things. Principal risk taking concentrated in one bank is a genuine depth story. Client flow routed through one broker’s plumbing is a different story wearing the same clothes. Either way, that much of the day’s flow moving through one counterparty’s book is itself worth naming.

Proprietary traders make up roughly a third of NSE’s normal turnover. Most stayed cautious about the new auction. The usual buyers, missing. Depth was not evenly distributed. It sat with one bank, and mostly absent from the desks that normally supply liquidity on an ordinary session.

Step three: a 3% price band turns a depth shortage into a hard stop. Not a slope. Roughly 60 stocks hit the auction’s 3% price limit that session. Sixty names, one afternoon. Eternal, Adani Enterprises and Reliance Industries were among the names swinging sharply. Once a stock’s clearing price would move more than 3% from its reference, the exchange does not let it move further inside that window. The imbalance has to resolve some other way, or carry into the next session.

The same August review removed Balkrishna Industries, SBI Cards and Astral from the index. Petajisto’s deletion data is why a rebalancing desk should care about outflows as much as inflows: minus 15.1% average impact, nearly double the addition side premium in his sample. A 20 minute auction concentrates exit flow the same way it concentrates entry flow. Same fragility, running the other direction, on the same clock.

Read more

Market maker vanna exposure, the Greek that decides whether dealer hedging dampens a selloff or force feeds it, just printed its steepest negative reading in two years, from a VIX nine points lower than the one that preceded the biggest one day volatility spike ever recorded.

Below the paid line:

• The full Black Scholes vanna calculation, six inputs shown, landing at a d1 of 1.19 and a vanna near -1.40.

• The two scenario BOJ playbook: the under two point non event case against the three point plus confirming case, sized against the current vanna extreme.

• The exact CBOE SKEW reading, 151.58 as of 4 September, that would make me stand down the whole trade if it reverses.

• The three point VIX threshold and the two week dated calendar that confirms or kills this call.

  

Photo: Wikimedia Commons · Public domain · via Wikimedia Commons

The consensus, stated the way its believers would state it

Ueda has telegraphed this hike for weeks. Swap markets price it in the 80s percent, which most desks treat as effectively decided. A textbook efficient market puts the information in the price before the meeting starts. On that reading, the meeting itself should pass quietly, because there is nothing left to learn from it. That is the standard case, and I want it at full strength before I argue with it. Nothing left to trade, on this reading.

There is a second leg, and it is the stronger one. The VIX sits at 14.53, near the low end of its 2026 range. The S&P 500 closed at 7,718.60 on 4 September, close to its 2026 highs. A calm market has, historically, gone on being calm for long stretches, and the base rate on any single quiet Tuesday resolving into a crisis is small. September itself carries a documented seasonal headwind, the weakest calendar month for US equities since 1950 (Chase, on S&P 500 seasonality) and the one where the VIX typically climbs off its summer lows, and that pattern shows up most years with no vanna story required. Trade every low VIX print as a warning and you spend years paying for insurance nothing collects on. Most Septembers, that’s the whole story.

Both legs describe the setup correctly. I think they draw the wrong conclusion from it, because the mechanism that actually moves the index sits one layer beneath the level everyone is reading. Wrong layer, right facts.

The variant view: vanna at a two year extreme, from a nine point lower VIX

My read is narrower than “a crash is coming.” The market has priced the BOJ’s decision correctly. What it hasn’t priced is how elastic its own reaction will be, because the structural condition that turned an unremarkable jobs miss into 2024’s largest one day spike is back. It carries more room to run this time than it did in 2024.

Risk.net reported on 28 August 2026 that market maker vanna exposure, a second order Greek measuring how an option’s volatility sensitivity shifts as the underlying moves, hit its steepest negative level in two years. No other outlet or vol analytics shop has independently corroborated that specific magnitude claim; it traces to this one paywalled report, and I flag that plainly. Repeating the headline elsewhere does not count as a second source. The newsroom’s own framing calls it an echo of August 2024, “when the VIX volatility index made its biggest ever intraday surge on a relatively muted stock fall.” I can’t reproduce a dealer’s book from public data, and I’ll say exactly where that limit sits below. One source, real claim, named limit.

What I can verify is the setup it’s landing on: a VIX that opened this cycle nearly nine points under where 2024’s spike began, and VIX futures open interest that climbed from 343,094 to 410,574 contracts in five weeks. That’s a 19.7% jump, and it arrived as a series of sharp steps over five weeks (CFTC). The level itself is not unusual on its own; open interest ran higher than today’s print twice already this year, 440,161 in May and 417,768 in August, with no crisis attached to either reading. What stands out is the shape of the last five weeks against the two years of dealer positioning Risk.net describes, not the raw count. Five weeks, one direction.

A positioning extreme sitting under a complacency priced index is the exact signature of the last two vol events this severe. The size of the coming shock doesn’t worry me much. What a shock of ordinary size does when it lands on a structurally short vanna book does.

The mechanism: why dealers are structurally short vanna

Start with why dealers carry this position at all. Their own client flow hands it to them. Garleanu, Pedersen and Poteshman’s 2009 study documents that end users, mostly institutions buying downside protection, hold a persistent net long position in S&P 500 index puts. Every long put an institution buys sits as a short put on a dealer’s book. Dealers cannot fully offset that inventory when the market structurally wants more protection than anyone wants to sell. Somebody always holds the other side.

A short put position is short vanna. Here is the plain version. Vanna tells you how much an option’s delta, its sensitivity to the underlying, moves when implied volatility moves. A dealer short vanna, watching volatility rise at the same moment the index falls (the well documented negative correlation between spot and vol, sometimes called the leverage effect), sees a hedge requirement that grows in the same direction as the price move. That is the whole mechanism. A hedge that grows toward the move does not slow it down. It adds fuel.

The causal chain deserves to be stated as a chain, because a compressed version of this is where most writing on the subject loses the reader. Four links, in the order the money actually moves.

1. Institutions buy downside protection continuously. Dealers absorb the resulting short put inventory because somebody has to take the other side. That inventory is structurally short vanna. Ordinary flow, permanent position.

2. A macro catalyst, here a BOJ decision priced in but still carrying surprise risk on magnitude and language, moves spot and implied vol together. This is the trigger, and it is not the cause. It’s the weather, never the fault line.

3. As vol rises into a falling tape, the dealer’s negative vanna position pushes the short put book’s effective delta further negative. The dealer needs more short exposure, or less long exposure, to stay hedged. The hedge moves with the pain.

4. The dealer sells into the decline to rebalance. That selling is a new market input the original catalyst never generated on its own. It is what turns a jobs report miss into a 42 point premarket VIX spike. Four links, one spring.

   

Every vol analytics shop that publishes on dealer flow, from SpotGamma’s gamma exposure work to the practitioner literature on charm and vanna more broadly, describes negative gamma and vanna positioning the same way. It makes hedging procyclical: dealers sell into weakness and buy into strength. That is amplification, stated plainly, and it is consensus among the practitioners who watch this for a living. Nobody outside the dealer desks can measure precisely how loaded the spring sits right now. That part remains a genuine unknown. Risk.net’s reporting puts it at a two year high. It stays unmeasurable from here.

The coverage has undersold one point worth adding. The BIS’s own postmortem on August 2024 attributes the amplification to procyclical deleveraging in the yen carry trade, roughly ¥40 trillion, about $250 billion, of on and off balance sheet positioning unwound inside days (BIS Bulletin 90), a figure the BIS itself flags as biased down by gaps in the underlying data, so the real unwind was probably larger than the headline number. Understated by the source’s own admission. That is a real and separate channel from options vanna, probably the larger dollar amount in motion. The two channels do not compete for the same explanation. They run in sequence: the carry unwind sells yen and the assets funded by it, that selling shows up in equity indices, and the equity move is what then meets whatever vanna position dealers are carrying. Both channels were loaded at once in 2024. My read is that two loaded channels together made that spike historic.

Read more

The popular arithmetic sets a 54% equity claim against 100% of consolidated debt. Correcting that halves the gap, and then a harder problem shows up underneath.

On 2 September, Bloomberg reported that Elliott Investment Management had built a position in Deutsche Telekom and wanted the company to drop a full combination with T-Mobile US in favour of larger buybacks. Reuters matched it the same day with its own source. Within hours the framing had set: more than 70% of Deutsche Telekom’s value is the American stake, Europe’s largest telecom business is therefore nearly free, and an activist has arrived to prise it out.

I spent two days rebuilding that arithmetic from the filings, and I got it wrong twice before I got it right. The popular version is wrong. My first correction was also wrong. What survives is a claim I can defend line by line, and it doesn’t need a peer table or a valuation opinion to stand up.

Deutsche Telekom owns controlling stakes in three other listed telecoms companies. All three trade in public markets every day. And the German business inside Deutsche Telekom is being valued at less than they are.

There is a film of this piece. It works the sum of the parts on screen, the minority-interest add-back, the lease double-count that produces a ratio Deutsche Telekom does not publish, and the thirty-second check that catches it, with the two SEC filings marked at the figures they carry.

Below the paid line: the 4.7x sum of the parts on one basis, the correction line by line with the two joint ventures the residual absorbs, the EUR 8.5bn segment lease split DT never discloses, the ratings notch a merger costs, and what would change my view.

The term the popular arithmetic drops

T-Mobile’s Q2 10-Q puts shares outstanding at 1,072,671,613 as of 17 July, of which Deutsche Telekom holds 582,065,875 directly, per its latest Schedule 13D/A. At the 2 September close of $187.30 that stake is $109.0bn, or €94.2bn at the ECB rate. Against a market capitalisation near €138.0bn, the stake is 68.2% of the parent, not “more than 70%”.

Subtract and you get about €43.8bn for everything outside America. Written that way it looks like the market is giving Germany away.

It isn’t, and the reason is a term the subtraction drops. Deutsche Telekom consolidates 100% of T-Mobile’s borrowings, because consolidation follows control, while its equity reflects only its 54% of T-Mobile’s equity. That comparison is a category error. You have to add back the roughly €79bn of minority interest DT consolidates but doesn’t own, and put net debt on both sides on one basis.

Do that and the non-US business comes out near 4.7x EBITDA after leases. Which is where I stopped, and where I was still wrong.

Two mistakes, because both are instructive

The first was leases. Deutsche Telekom’s H1 2026 report publishes net debt two ways: €138,387m including lease liabilities, €104,352m excluding them. I took the figure that includes leases and divided by EBITDA after leases. EBITDA AL is already struck after lease cost, so that charges for leases twice. I then ran a “second, independent” calculation that agreed to the euro and treated the agreement as confirmation. It was the same equation rearranged. A rearrangement confirms nothing.

The second was the residual, and it’s the one that matters. My €72.5bn of non-US enterprise value is what’s left after a €274bn subtraction, and a residual absorbs everything you failed to model.

Two things sit in it that shouldn’t. Deutsche Telekom’s 49% of GD Towers and 50% of GlasfaserPlus are joint ventures carried at equity. Their value sits inside DT’s market cap, and therefore inside my residual, but their earnings are not in my €15.4bn denominator. DT prints a table of zeros in its annual report to make the point: since February 2023, GD Towers “has no longer been part of the Group.” DT’s own Level 3 fair value marks are €7.5bn and €1.1bn.

Pulling in the other direction, DT consolidates 100% of the EBITDA of three listed European subsidiaries whose minorities are worth about €6.6bn at market, and those minorities belong in enterprise value.

Strip the stakes without the minority added back and the pure operating business is nearer 4.15x.

I want to be clear that I am not confident to two decimal places here, and nobody should be. What I am confident about is the direction: every correction I have been able to make to the naive number has moved it the same way, and the corrections that would move it back are ones I looked for and could not find.

The comparison that needs no opinion

Here’s why I stopped worrying about peer tables. There is no free data source that publishes EV/EBITDAaL for European telecoms, and the same company can span 40% across four providers on one day. Any argument resting on a “European telco average” is averaging across a 2.7x spread.

So use Deutsche Telekom’s own subsidiaries instead. DT is the parent of four listed companies, not one. Alongside T-Mobile US it controls OTE in Greece, Magyar Telekom in Hungary and Hrvatski Telekom in Croatia. All three are quoted daily. All three carry almost no net debt, which removes the leverage argument entirely. And together they are roughly half the Europe segment’s earnings.

When I pulled their market prices I got this: OTE at about 5.6x forward and 6.4x trailing. Magyar at 5.6x and 6.8x. Hrvatski Telekom at 7.0x and 6.9x.

The non-US block blends to roughly 4.6x. Since half of it is marked in the open market between 5.6x and 7.0x, the other half has to be lower. Telekom Deutschland, which produces a 41% EBITDA after leases margin and takes 44% of all German telecom revenue, per the Bundesnetzagentur, is implied at less than its Hungarian and Croatian subsidiaries.

That argument needs no convention, no peer set and no judgement from me about what a fair multiple looks like. It is market prices on both sides of the comparison, which is why I trust it more than the number I spent two days deriving.

I’d still hold it loosely. The stub is a levered residual, so a $10bn move in T-Mobile’s net debt shifts it half a turn. But the direction survives every version I’ve run.

The ten million shares that show how this structure really works

There is a transaction buried in Amendment No. 35 that explains the whole architecture better than any ownership percentage.

SoftBank’s residual holding in T-Mobile is now exactly ten million shares. On 19 March three agreements were executed at once. A novation moved something called the SB-DT 2029 Call Option to a SoftBank entity called Signal Bright. Signal Bright then entered a variable prepaid forward sale contract with an unaffiliated dealer over those shares, pledging them as security. And Deutsche Telekom took a right of first refusal over the pledged stock.

A variable prepaid forward contract lets a holder take cash now against shares delivered later. It is the standard way to monetise a concentrated position without selling outright. So SoftBank gets its money, and I think that is the whole point of the structure.

The votes, though, do not travel with it, and that asymmetry is what I want you to notice. The filing is explicit that Deutsche Telekom “will continue to have the right to direct the vote of the Pledged VPF Shares under the Proxy.” Economics to the dealer, votes to the controller.

One caution against reading too much into it, which I’ll make because I nearly did. The 2029 call option’s strike is the previous day’s closing price. It can never be in the money. It functions as an access and blocking right, and carries no intrinsic value at all. What the arrangement shows is how control gets assembled here: out of contracts instead of purchases, durable enough to survive a counterparty selling out from under it.

Read more

Two Sigma’s Ownership Trusts were built so that no single life event at either founder could touch who controls the $75 billion firm, and John Overdeck’s own sworn testimony in his New Jersey divorce trial is now the evidence a court would need to treat that same trust interest as illiquid enough to discount, under a legal standard that otherwise forbids exactly that discount.

Photo: Don Ramey Logan · CC BY 4.0, resized · via Wikimedia Commons

A billionaire divorce, irrelevant to the fund

Read the coverage this week and the story is a celebrity divorce wearing a hedge fund’s clothes. Two billionaires. No prenup. A $6.2 billion number that makes headlines on its own.

That reading isn’t naive. Family offices see marital splits among the ultra wealthy constantly, and the standard playbook exists because it usually works. The business keeps running. The founder writes a check or a note. A few years later nobody outside the family remembers the number. That’s the base rate.

Two Sigma’s own numbers back that up. The firm is managing more money than it ever has, a record north of $75 billion, per Hedgeweek’s reporting on the 2023 investor departure, which the firm replaced and then grew past. A quant manager whose systems and headcount are intact doesn’t need Overdeck’s personal balance sheet to keep functioning.

On this view, the right professional reaction to “hedge fund cofounder’s divorce” is the same reaction a PM has to any billionaire’s marital news. Interesting. Irrelevant to the fund.

Overdeck’s testimony is doing legal work

I think that reading skips the one detail that actually matters to anyone who allocates to, competes with, or works inside a multi strategy manager. Overdeck is not describing a personal liquidity problem. He’s describing a control problem, under oath, in a jurisdiction whose leading precedent says his kind of interest should not get a discount for exactly the reason he’s describing.

That is a specific legal maneuver. Two Sigma’s own regulatory filings are what make the maneuver credible.

New Jersey values marital business interests at fair value, a specific legal standard distinct from fair market value. The state’s default rule, from Brown v. Brown (App. Div. 2002), holds that marketability and minority discounts do not apply absent “extraordinary circumstances”: “lack of liquidity does not affect the fair value of the minority interest,” full stop, unless the case clears that bar. Brown never defined what would.

Overdeck’s testimony reads like an attempt to clear it. Family law commentators have argued, following a 2022 state Supreme Court case about a closely held business buyout, that “extraordinary circumstances” should reach fact patterns like a buy sell agreement limiting what an owner can receive, or an owner who lacks real control over compensation and capital calls.

I went looking for the actual case behind that argument. It isn’t what it’s cited as. I want to be precise about what that 2022 case, Sipko v. Koger, actually is and is not: it is a shareholder oppression dispute between family members over two closely held companies, not a divorce case, and its own outcome ran the other way. The Court denied a discount because the controlling party had acted in bad faith to deplete the company’s value, the opposite fact pattern from Overdeck’s. The buy sell and compensation control framework is the commentary of one family law practitioner writing after that case, not a holding any court has applied to a fact pattern like this one. No court has yet found “extraordinary circumstances” in a case resembling Overdeck’s.

What his testimony has going for it is narrower than “new precedent, directly on point.” It reads like a serious attempt to build the record Brown itself said would matter, using the one argument NJ’s family bar has actually been making. I cannot prove that is his legal team’s deliberate strategy. I can show you it is a fit with the argument, not with a decided case.

The mechanism: a trust built to protect control becomes the exhibit that explains why it might not

Start with what Two Sigma actually built. Per Two Sigma Investor Solutions LP’s own Form ADV Part 2A brochure, an SEC-registered Two Sigma affiliate, public and current as of the March 2025 update, trusts established by Overdeck and Siegel, called the Ownership Trusts in the filing, are “the principal owners of the Adviser.” Those Ownership Trusts “indirectly effectively control” TSM, the general partner, and the advisory business itself.

That structure isn’t decoration. Routing founder control through a trust is a standard way large private managers wall off the firm from a single founder’s estate, incapacity, or a life event like a divorce. It’s the same logic that puts a founder’s operating company shares into a family trust or a voting trust well before anyone expects a fight: name the vehicle as the owner, and no single life event touching the individual is supposed to touch the vehicle. My read is that the design intent here was precisely what Overdeck is now testifying against: making sure nothing that happens to John Overdeck the person can, by itself, touch who runs the business.

Layered on top sits a two person Management Committee, one seat per Co-Chairman. Each Co-Chairman can unilaterally remove his own appointee. That’s 50/50 by construction, with no tiebreaker. I’d call that the single most fragile governance shape a manager this size can run: it works exactly as long as two people keep agreeing, and it has no built in answer for the day they stop. No committee. No board. No vote.

The firm’s leadership actually moved through this exact structure recently, and the dates matter. On September 30, 2024, Overdeck and Siegel both stepped back from the Management Committee itself, each naming a designee in his place. As of March 31, 2025, Overdeck reversed course and reclaimed his own seat, replacing his designee. A structure this sensitive to who occupies two chairs is a structure where the identity of the equity holder behind each chair isn’t a footnote.

Here’s the chain, in the order the risk actually moves. Four links. Money has to move through all four before the story is real, not hypothetical.

One. A New Jersey court values Overdeck’s trust interest for equitable distribution, somewhere between his $4.9bn and her $6.2bn.

Two. The court orders a payment Overdeck can’t fully cover from outside liquid assets alone. His own testimony is that this is a real possibility above his $723M offer.

Three. Meeting that order means monetizing part of an Ownership Trust interest that was never designed to be sold in pieces, through a pledge, a secondary sale, or a structured payout that still changes who holds an economic claim on the trust.

Four. Because the trusts are what “indirectly effectively control” the Adviser, any change to what sits inside Overdeck’s trust is a change to the mechanism that was supposed to keep 50/50 parity intact.

Each link is disclosed somewhere on the record. The ADV brochure supplies links one and four. His testimony supplies two and three. No single document states the whole chain, which is why it reads as a divorce story instead of a governance story to anyone reading only one source.

The tell that predates the trial by three years

The independent tell is the ADV brochure itself, and it predates the trial. Two Sigma’s Item 4 risk factors, filed under the heading “Certain Risks Associated with Management and Governance Challenges,” disclose that the Management Committee “has been unable to reach agreement on a number of topics,” specifically naming C-level role definitions, “organizational design and management structure,” “corporate governance and oversight matters,” and succession planning.

That sentence sits in a document the SEC can act against if it’s false or misleading. That’s a materially higher bar than anything said in a divorce deposition. I wouldn’t weight Overdeck’s trial testimony alone as proof the control risk is real. I weight it alongside a filing the firm had every incentive to soften and didn’t, filed year after year, long before a divorce trial gave anyone outside the firm a reason to read it closely. Nobody was reading it for this. It’s also not the first time Two Sigma’s own regulatory record has cut against its public reputation: I wrote up its $90 million SEC fine earlier this year, on a separate compliance failure the firm itself disclosed.

The same two week window in August produced two more data points, and both firm the mechanism up. On August 17, entities linked to Siegel’s family filed arbitration against Overdeck and co-CEO Carter Lyons, alleging undisclosed conflicts tied to a former employee’s severance and deferred compensation claim. Hedgeweek reported the dispute escalated all the way to formal arbitration, something the Management Committee itself was evidently unable to resolve internally.

The clearest account of the second dispute’s mechanics isn’t in any press report. It’s in Two Sigma’s own Form ADV, filed with the SEC on March 31, 2026. Siegel designated Seth Platt his Executive Manager and co-CEO on March 16, 2026, replacing Scott Hoffman, who Hedgeweek reported had resigned citing “continuing governance difficulties.” Platt then moved to terminate co-CEO Carter Lyons, arguing Lyons had undermined his authority. Overdeck called that move “imprudent and baseless” and placed the termination into dispute resolution, meaning it cannot take effect without Management Committee agreement or an arbitrator’s ruling. Whether an Executive Manager appointment even confers the co-CEO title at all, the filing states plainly, “is in dispute.”

That is Two Sigma telling the SEC, in its own words, that one Co-Chairman’s designated lieutenant tried to fire the other Co-Chairman’s designated lieutenant, and the firm’s own governance apparatus could not settle it without outside intervention. The same regulatory filing names the underlying ownership structure precisely: the “John A. Overdeck Revocable Trust” and the “David M. Siegel Revocable Trust,” both Delaware entities, both Limited Partners in the 25-to-50-percent band, each founder listed as trustee of his own trust. As of February 2026, a third party, Thomas Rafferty, was added as an additional trustee on Siegel’s trust. Overdeck’s trust carries no such addition on the same filing. I can’t tell you what that difference means. I can tell you it’s the kind of structural fact this system was supposedly designed to make irrelevant to any single person’s circumstances, and it isn’t. One trust got a second trustee. One didn’t.

Three separate legal proceedings. Two founders. Sixteen days. That’s what a firm running its command structure on ad hoc litigation looks like, and it’s happening at the exact moment a family court is being asked to decide how much one founder’s economic interest is worth. The Platt-Lyons succession fight itself already has a second life: I tracked how fast the public record moved versus how slowly an actual institutional allocator reacted once the story broke.

What “monetizing part of an illiquid trust interest” would actually look like is worth spelling out, because it isn’t a hypothetical this industry has never seen. Founder and GP-level stakes at private managers get sold every year to specialist buyers: Petershill, and Blue Owl’s GP Strategic Capital (the business Dyal Capital Partners became after its 2021 merger). A founder’s economic interest in a fund manager is illiquid by design, and a market exists to price that illiquidity anyway. A structured minority sale of a slice of Overdeck’s trust interest, or a margin loan collateralized by it, are the two realistic mechanisms for closing a multi billion dollar gap without selling the whole interest. Either one still changes who has a claim on the trust that “indirectly effectively controls” Two Sigma, even if Overdeck’s own voting seat never formally moves. The control risk doesn’t require a headline sale to be real. It only requires a lender or a minority buyer with a seat at the table that didn’t exist before.

The timeline: this was never a sudden story

Lay out the dates in order and the divorce trial stops looking like the origin of the problem. It starts looking like the latest entry in a ledger that has been running for three years.

March 31, 2023. Two Sigma first disclosed the Overdeck-Siegel rift to regulators as a business risk, in a Form ADV update Fortune reported on the same day it became public. The filing’s own language warned that “if such disagreement were to continue, the adviser’s ability to achieve client mandates could be impacted over time.” Around the same window, a major investor exited the platform citing that governance concern. The firm replaced the capital. The disclosure did not go away. It is still there, almost word for word, in the March 2025 update I read for this piece. Two years of an unresolved risk factor is not a disclosure. It is a standing admission that nobody inside the firm has found a structural fix.

September 30, 2024. Overdeck and Siegel both stepped back from the two person Management Committee itself, each naming a designee, and became Co-Chairmen of the general partner instead. The idea, I would guess, was to put daily governance in less personally entangled hands while the founders kept ultimate control through their trusts.

March 31, 2025. Overdeck reversed that, rejoining the Management Committee directly and replacing his own designee. Whatever the designee arrangement was supposed to fix, Overdeck judged it insufficient inside six months.

August 17 and 18, 2026. Two arbitrations, a day apart in filing terms and a world apart in subject: Siegel-linked entities accusing Overdeck and co-CEO Carter Lyons of undisclosed conflicts, and Seth Platt disputing whether his own committee seat carries the co-CEO title he believes it does.

September 2, 2026. Overdeck testifies that preserving voting parity with Siegel is a “principal concern” in his own divorce, the same figures and quote Bloomberg’s original report carries, independently confirmed here.

My read of that sequence: every governance fix Two Sigma has tried since 2023 has lasted less than two years before someone inside the firm needed a court or an arbitrator to settle it instead. Designee committee seats. A Co-Chairman layer. A title clarified by memo, never formalized in a document. None of it held. Every fix bought months, not years. A structure that keeps needing outside adjudicators is not stable. It has been quietly outsourcing its own governance function to litigation, and the divorce trial is simply the first instance visible to people who do not work there.

What any LP in a founder controlled fund should actually check

This is a design pattern common to nearly every founder led multi strategy manager, and Two Sigma’s public filings happen to be unusually legible about it right now. Three checks translate directly to any fund where control sits with one or two named individuals rather than an institutional board.

Read the ADV Part 2A’s risk factors section for governance language, not just performance and conflicts disclosures. Most LPs skim Item 4 for AUM and skip straight to fees. Two Sigma’s own “Certain Risks Associated with Management and Governance Challenges” heading has been sitting in plain text, publicly filed, for anyone who reads past the first page.

Ask directly whether founder economic interests sit in a trust, a direct partnership stake, or something else, and whether that vehicle has ever been tested by a life event. A trust that has never faced a forced valuation is an assumption. Nothing more. Two Sigma’s is being tested in real time, in public, right now.

Check whether governance depends on a fixed number of named seats with no tiebreaker, versus a board with independent members or a supermajority mechanism. A 50/50 committee with unilateral removal rights per seat, which is what Two Sigma discloses, has no structural way to break a genuine deadlock short of the two principals agreeing or an outside body forcing the question. That’s the exact condition under which one founder’s personal legal exposure becomes the whole firm’s problem.

None of those three checks require anything beyond documents every LP is already entitled to receive. Most never ask. That’s the gap, not the filing.

I ran this exact three clause check end to end on Two Sigma’s own filing, with the precise document retrieval path, the CRD lookup, and the dated trigger this specific case turns on: Two Sigma’s Own SEC Filing Shows Exactly How a $4.9bn Ownership Trust Works.

Strategic testimony, or a real balance sheet event?

The rival explanation deserves its full weight, because it is the one a skeptical family law practitioner would raise first. Spouses routinely overstate illiquidity and control risk specifically to argue the business interest down. It is close to the single most common maneuver in a high net worth divorce, and Overdeck’s counsel would be negligent not to make it.

Under that reading, “maintaining voting parity is a principal concern” is a line written for a judge. The ADV language is just risk factor boilerplate that every multi strategy manager with founder tension eventually files.

What separates the two readings is timing and audience. A risk factor written for a securities regulator, in a document that predates the trial by years and carries its own legal exposure for misstatement, is a different kind of evidence than testimony written for a family court judge weighing a settlement.

If the Management Committee gridlock were purely a talking point invented for this trial, it would not already be sitting in a 2025 client facing brochure describing succession and governance disagreements the firm had every reason to minimize. Two Sigma’s compliance and legal teams reviewed and filed that language before anyone knew a divorce trial would ever quote it back.

The ADV disclosure does not prove the control risk framing is honest. It proves the underlying gridlock is real and independently documented. That is the one fact a purely strategic reading of his testimony cannot explain away. The paper trail predates the motive.

Four things I can’t verify from public documents

I can’t verify the actual terms of the Ownership Trusts. The buy sell provisions, whether either founder holds a contractual right of first refusal on the other’s interest, whether the trust documents contain a specific divorce triggered transfer clause: none of that is public. The ADV brochure describes the control effect, not the trust mechanics that produce it.

I also can’t confirm the New Jersey court will accept a forced sale theory at all. Overdeck’s testimony argues for it, but no ruling on valuation methodology has been made public, and the trial doesn’t resume until October.

And the AUM record itself is a genuine confound. If the market believed governance risk at Two Sigma were acute, a multi strategy manager that depends on continuous capital raising would be the first place it showed up. It hasn’t shown up there yet. The market isn’t pricing this.

A fourth complication, and it cuts against my generous reading of the trust structure’s design intent. Laura Overdeck has a separate, still live lawsuit against the law firm Seward & Kissel, reported by The Wealth Advisor, alleging the firm moved several billion dollars of marital assets into new Wyoming trusts in 2018 without disclosing that the new terms would strip her beneficiary status the moment either spouse filed for divorce. Those are different vehicles from the Ownership Trust this piece is about, so the allegation doesn’t touch the ADV facts above. But it means I can’t take “the trust structure was designed as neutral governance protection” as a given about this particular person: his family has a separately litigated history around trust vehicles and divorce triggered terms, and a reader should weigh that when judging how much good faith to extend to the design intent reading. Good design and good faith are not the same claim.

I also can’t tell you how contested the $6.2bn figure currently is inside the actual courtroom: Laura Overdeck has been barred from testifying and from calling her own valuation expert at trial, a sanction Judge Buechler imposed for discovery misconduct: accessing Overdeck’s private computer and mailbox, and photographing emails before deleting the evidence. The $6.2bn number is on the record, but the witness who produced it may never take the stand. A number nobody can defend live is a number, not a fact yet.

That gap, between a record asset base and three simultaneous legal proceedings over control, is real. I’m not going to resolve it by assuming the quieter number is the one that’s wrong.

What would change this view

Watch three things, each dated and public.

The underwriting question for every LP in a founder controlled fund

The professional question this week isn’t whether Two Sigma survives a divorce. At $75 billion and growing, it almost certainly does.

The real question is whether “control via trust” holds up the first time a court outside the firm’s own governance documents gets to test it, or whether it only ever worked because nobody had tried.

My read: the ADV brochure is the tell that the Management Committee’s gridlock was real before any of this reached a courtroom. The control question Overdeck is litigating isn’t manufactured for the case. It’s the same gap the firm has been disclosing to its own clients since March 2023, arriving at a different address.

Which side of that gap would you underwrite? The trust that has held through more than three years of documented gridlock, or the trust that has never yet faced a court order it couldn’t simply absorb.

The Decision-Grade Version

This piece is complete on its own. The thesis, the evidence, and what would kill the view are all above, and nothing was held back to sell you a next step.

The Patreon note is a separate piece of work rather than a deeper cut of this article. It takes the exact SEC filing method this piece runs on Two Sigma, i.e. the three clause Schedule A and Item 4 checklist, and writes it as an instrument any reader can run themselves on any founder controlled fund they hold. Written for people who do their own due diligence.

→ Read the Two Sigma governance checklist note

→ Or join the Patreon community for every note

More research like this: YouTube · LinkedIn

Cover photograph: Tdorante10, CC BY-SA 4.0, via Wikimedia Commons.

A hedge fund’s own compliance paperwork disclosed a new pod’s brand and website before any press outlet reported the hire.

The figures behind all of it, pulled straight from the filings:

Below the paid line:

• The full timeline: July 31 (data as of), then August 19 (form date), then September 3 (Hedgeweek)

• The exact Schedule D miscellaneous language ExodusPoint filed, quoted verbatim, plus the CIK, CRD and 801 numbers to pull it yourself

• ExodusPoint’s June 30, 2026 13F snapshot: $15.53bn reported long book, 1,510 positions, the options split

• The $2bn Abu Dhabi backed raise this pod draws against, and what fraction of it $1bn+ represents

• WestWind Asset Management: ExodusPoint’s only other prior other name, the 2021 loss event tied to its era, and why its ADV entry is still a live check today

• A dated, falsifiable trigger: what to watch for at ExodusPoint’s next annual amendment, due by March 2027

The consensus: a hedge fund hire is a press event

Trade coverage of the multi strategy pod shops treats a new portfolio manager’s arrival the way it treats an earnings beat. A fact exists the day it gets reported. Hedgeweek’s story this morning follows the standard shape. A senior macro trader with a name, Ben Atlas, formerly of BlueCrest, leaves one platform and lands at another with a headline allocation of “more than $1bn.” The story runs. The aggregators pick it up. LinkedIn does the rest. Simple. Familiar.

That reading captures something real. It’s where the information turns usable for a reader. State it at its strongest: trade journalism is the primary record for personnel moves at private funds, because private funds don’t issue an 8-K when they hire someone. There’s no earnings call. There’s no proxy statement. A private multi manager platform’s staffing stays opaque, by design, until somebody with sourcing inside the industry writes it down.

One correction before the variant view: this isn’t the first time someone read an ADV filing

The consensus above has already moved once, and I want to be honest about where. Mining Form ADV for staffing news isn’t unknown. eFinancialCareers already tracks headcount, year over year, at Millennium, Point72 and Jain Global, straight from ADV’s own employee count item, published before those firms say a word about it themselves. That practice is real, and it isn’t what I found. Headcount is one number, filed once a year, aggregated across an entire firm. Different field. Different clock. Schedule D’s other business names section is a different field entirely, filed the moment it changes, naming one pod at a time. Nobody appears to be reading that field yet.

I don’t think the press comes first, not here. I can point to the document that says so.

The variant view: the SEC already had it

ExodusPoint Capital Management, LP is registered with the SEC as an investment adviser under CRD number 294156, SEC file number 801-112901. I expected its filing history, if it showed anything at all, to read as routine housekeeping. I pulled its SEC investment adviser record directly, on the IAPD system the SEC runs specifically for adviser filings, separate from EDGAR, to check. It did not read as housekeeping. The most recent filing on record is not the annual amendment every adviser owes once a year. It is an “Other-Than-Annual Amendment, All Sections,” and the generated form itself carries the date 8/19/2026 5:50:30 PM. A specific name. A specific date. Not housekeeping.

Schedule D’s miscellaneous section exists so an adviser can explain, in its own words, what an amendment actually changed. Here is ExodusPoint’s own ADV brochure, quoted verbatim:

“This Other-than-Annual Amendment reflects the addition of Atlas Capital in Schedule D, Section 1.B and the associated website for Atlas Capital in Schedule D, Section 1.I. Further, Item 5.B(3), Schedule D(1.F), and Schedule R(4.A) were updated as of July 31, 2026.”

Section 1.B of that filing lists “Name: ATLAS CAPITAL,” registered in New York. I checked the actual jurisdiction grid rather than trusting the label list a plain text export throws back: of the 55 state and territory boxes on that page, exactly one is checked. Section 1.I lists its website, atlas-capital.com, next to ExodusPoint’s own site and its LinkedIn and Glassdoor pages. That’s the same name AlphaMaven’s aggregation of the Ben Atlas story carries for the pod he’s said to be launching later this month, though AlphaMaven’s own byline field credits Hedgeweek, so this is one wire syndicated twice, not two separate confirmations. Weigh it that way rather than as double corroboration.

Three dates, one document, one wire story. July 31 is when the underlying change was already true. August 19 is when ExodusPoint told the SEC. September 3 is when Bloomberg told the market, with Hedgeweek and AlphaMaven both relaying the same reporting. The compliance record beat that story to this hire by at least fifteen days. Measured against the date the underlying change was reported true, it beat it by more than a month. Paper first. Wire second.

I checked the obvious falsifier before writing a word of this. Is Atlas Capital a name ExodusPoint has used for something unconnected to Ben Atlas? I found no other candidate. The name is new to this amendment. Nothing in ExodusPoint’s filing history before August 19 carries it. No prior use. The falsifier does not land. I also checked the domain live while drafting: atlas-capital.com doesn’t currently resolve. A registered brand, waiting on its own launch.

The mechanism: why the paperwork runs ahead

The credit belongs to SEC Rule 204-1, a function of how the rule actually works, no leak and no clever detective work on my part.

Registered advisers file Form ADV’s annual amendment once a year, within 90 days of fiscal year end. The SEC’s own General Instructions require something else too. Additional amendments, filed promptly, whenever information reported under Item 1 becomes inaccurate. Other business names live inside Item 1, in Section 1.B. The instructions carry real teeth: “failure to update your Form ADV, as required by this instruction, is a violation of SEC rules or similar state rules and could lead to your registration being revoked.” Real teeth. Compliance does not wait for a press cycle.

A firm cannot wait for its annual cycle to disclose a new trading name once that name is actually in use for advisory business. That’s the exact phrase the form uses: Section 1.B asks for any additional names under which the adviser conducts its advisory business. Registering Atlas Capital is not a marketing flourish. It is ExodusPoint telling the SEC it is, or will imminently be, legally doing business under that name. A regulatory act with a deadline attached. Nothing softer than that.

A press announcement runs on an entirely different clock. It waits for the hire to close. It waits for both sides to agree on language. It waits for a trade reporter’s sourcing to come together. None of that carries a regulatory deadline. The compliance filing has one. The news story doesn’t. That asymmetry is the whole mechanism, and it isn’t personal. Nobody at ExodusPoint is trying to bury this in EDGAR. The filing exists because the rule requires it, on a timeline the rule sets, whatever the trade press decides to do with the story later.

I turned the actual checking process into its own walkthrough, the exact CRD lookup, the PDF field to read, and the jurisdiction checkbox mistake that cost me a correction inside this very piece. The Five-Minute ADV Check has all of it, so you can run this same check on any adviser yourself, not just this one.

Read more

Two Sigma’s Absolute Return Enhanced Fund gained 15% in 2025 and is up 12.4% this year. In June, New Mexico’s Public Employees Retirement Association pulled roughly $100 million out of it anyway.

PERA’s own explanation, as reported by Hedgeweek, was that organizational issues at the firm had created “too much uncertainty, with no clear timetable for resolving the internal conflict.” The pension said this. It said so even as it acknowledged the strategy had “outperformed both the pension’s benchmark and comparable funds.” That’s a $100 million redemption from a fund that was working, on grounds that had nothing to do with the fund working.

I don’t think that’s the interesting part. Institutional investors pulling money over governance risk instead of performance risk is a known move, documented for decades in fund formation practice. What’s interesting is that the governance risk here was never obscure. It was on Bloomberg. The gap isn’t information. It’s decision speed.

Here is the short version of how I got there. I expected to find a hidden disclosure, a fact PERA should have caught and did not. It wasn’t hidden at all. The method was simple: pull the actual filing, pull the actual press dates, and put them on one timeline. The outcome flipped the story on its head.

The consensus: fast public news gets a fast institutional reaction

Ask an allocator what happens when a manager’s leadership structure blows up in the trade press, not in a quiet filing nobody reads but in an actual Bloomberg headline, and the working assumption is that a professional monitoring program reacts inside days, not months. That’s the whole pitch of an active manager monitoring program: pay a team or a consultant to watch the wire, and when something material breaks, the institution moves faster than a retail investor reading the same story a week later.

I think that’s a reasonable baseline, and it makes New Mexico’s timeline the thing that needs explaining, not excusing. PERA is not a small, understaffed plan. It runs an active monitoring program, by its own account. The news wasn’t buried. It named the two men, the disputed title, and the arbitration in the first 48 hours. Nothing about it was quiet.

The variant view: the delay sat between the headline and the wire transfer

Two Sigma Investments, LP filed its 2025 Annual Amendment to Form ADV on March 31, 2026, signed by James G. Hein, Jr. Buried in Schedule D’s free text Miscellaneous section, in the firm’s own words, is this:

“As of March 16, 2026, Dr. Siegel designated Seth Platt as his Executive Manager and as co-CEO. There is disagreement about whether, when designated, an Executive Manager becomes a co-CEO or whether the appointment of a co-CEO requires approval by both members of the Management Committee... Mr. Platt has moved to terminate Mr. Lyons given his view that Mr. Lyons undermined his authority as Executive Manager. Mr. Overdeck believes Mr. Platt’s actions in this regard are imprudent and baseless and has placed the termination into dispute resolution, which means that the termination cannot take effect without Management Committee agreement or resolution through arbitration.”

That’s a signed, dated, federally filed statement that the governance structure built around Two Sigma’s two founders had broken down to the point where one side was trying to fire the other side’s designated co-CEO. It reached the trade press almost immediately. One day, not five months. Bloomberg’s piece on Hoffman’s resignation ran April 1, and Hedgeweek’s followed April 2, both reporting the Platt/co-CEO ambiguity and the move against Carter Lyons, citing the same March 31 filing this article quotes (Bloomberg; Hedgeweek).

Two Sigma’s disclosure to press pipeline, on this filing, worked about as fast as institutional disclosure ever works. The interesting number isn’t how long the market took to notice. It’s how long PERA took to act once it had.

Separately, and later, a second and different set of facts became public: the Siegel family’s arbitration referral over a former employee’s severance claim, and John Overdeck’s own account, given in his divorce trial, that the firm had lost “a major investor” once before. That investor withdrew in 2023, according to a person familiar with the matter cited by Hedgeweek (Hedgeweek). Those specifics didn’t surface until Bloomberg and Hedgeweek reported them around August 17-27 (Hedgeweek; Bloomberg). That part genuinely lagged. The April news, the one that explains PERA’s stated rationale, did not.

Where the two founders’ structure actually came from

This did not start as a fight between two people with no formal roles. In August 2024, Overdeck and Siegel both stepped back from the Co-CEO title they had held since founding the firm and installed Carter Lyons and Scott Hoffman as Co-CEOs, effective September 30, 2024 (Two Sigma). One structure, two men sharing one title. That arrangement was the firm’s own governance fix for exactly the kind of stalemate risk a founder controlled shop always carries, and it held for about eighteen months.

It broke in two directions at once this year. Overdeck returned to his own seat on the management committee, which is what Hoffman later cited when he resigned, describing “ongoing governance challenges” that followed directly from that return (Hedgeweek). Siegel then named Seth Platt. Platt would replace Hoffman. That triggered the exact dispute Two Sigma disclosed in its March ADV filing: whether naming a replacement Executive Manager automatically confers the Co-CEO title, or requires sign off from both sides of a committee that, by this point, could not agree on much. Platt then moved to fire Lyons. He moved fast. Overdeck blocked it. He pushed the whole question into arbitration. The two founder veto structure had failed. At the one thing it existed to prevent.

Read that sequence next to the ADV language above and the disclosure was not vague. It named both principals, dated the disagreement, and stated plainly that a live termination fight could not resolve without either committee agreement or arbitration. Bloomberg had the same facts within a day. Same day, same story.

Peer multi strategy managers price this exact structural risk in how they lock up capital. D.E. Shaw, Millennium and Citadel lock up cash for years; Rokos chose the opposite. A founder controlled shop with a fragile succession fix is precisely the kind of manager a longer lock up is meant to protect against a fast exit like PERA’s.

Lay the dates out in order and the sequence reads clean. September 2024: Lyons and Hoffman become Co-CEOs. Early 2026: Overdeck reclaims his committee seat. March 16: Siegel names Platt. March 31: the ADV filing discloses the fight. April 1 and 2: Bloomberg and Hedgeweek both report it. Then a long stretch with no new public fact. Sometime in June: PERA redeems. The story moved fast. The money moved slow.

The mechanism: public disclosure has one clock, institutional redemption has another

A governance blowup moving through an allocator’s process runs on two separate clocks, and conflating them is what made my first read of this story wrong.

Clock one: how fast the information becomes actionable. Here it was fast. The ADV filing posted March 31. Bloomberg and Hedgeweek both had the story by April 2. They cited the filing directly. Two Sigma did not bury this in language nobody parses; a fund’s own free text disclosure requirement, plus a wire service reading EDGAR the way wire services do, moved this from private dispute to public headline in under 48 hours.

Clock two: how fast an allocator can actually redeem, even once it decides to. This is the clock that matters here, and hedge fund redemption terms are not instant by design. Industry-standard terms permit redemptions on a quarterly basis, typically on 30 to 90 days’ prior notice to the manager, with proceeds paid out within 30 days after that (Proskauer). Say Two Sigma’s Absolute Return Enhanced Fund runs anywhere near that standard. An LP that decided to redeem the day the news broke, April 2, could easily not see the money move until May or June under the fund’s own mechanical terms. That holds independent of how fast or slow PERA itself was to decide.

I cannot confirm Two Sigma’s specific notice period. Not for this fund. That is a real gap, and it is the one place this piece’s mechanism claim is a plausible industry standard inference rather than a verified fact. What can be said is that the gap between the April news and the June redemption, somewhere around nine to thirteen weeks depending on the exact June date, is at least consistent with a quarterly redemption structure carrying a 60-to-90-day notice window, layered on top of whatever time PERA’s own investment committee needed to review the news, decide, and file the notice in the first place. Both clocks run inside that gap. Neither one alone explains it.

A tighter version of this same mechanism shows up wherever a fund has formalized its own version of the second clock. Blue Owl’s redemption gate costs the fund itself $12 million a year to run, precisely because the gate is the second clock, priced and disclosed rather than left to an allocator’s own committee calendar.

That’s the finding, reframed from where I started: the redemption right was never hard to find, and the news was never hard to find either. What actually took time was moving through the machinery, on both sides, that turns a headline into a wire transfer.

I will name the version of this I cannot independently verify, because I think it matters more than it first looks. At least one compliance vendor, Dasseti, markets an AI tool that claims to read Form ADV’s narrative text directly and flag conflicts and risk language, not just the numeric fields. Whether PERA’s own monitoring stack uses anything like it is not knowable from here. Neither is whether it flagged this specific paragraph. It does not change the finding either way. Even a tool that read Schedule D the moment it posted on March 31 would have surfaced the same facts Bloomberg published on April 1. At most, it would have given PERA a one day head start over the wire services. It would not have produced the weeks of lead time the redemption gap implies. The bottleneck sits downstream of detection, wherever detection happens.

Two clocks explain the shape of the gap. They don’t tell you what your own number would be. I built a five input audit: trigger to filed lag, filed to press lag, your fund’s own notice period, your own committee’s cadence, and the compounded sum. It’s worked through Two Sigma’s real filing as the template, in a $5 institutional note on Patreon. It’s the number almost no allocator has written down before they need it.

A firm whose internal awareness has outpaced its disclosures before

There is a second, older instance of this pattern at Two Sigma, on a completely different matter. It is worth naming because it is a more extreme version of the same shape: something true inside the firm for years before it became public.

In January 2025 the firm settled SEC charges for failing to fix known vulnerabilities in its investment models across roughly four years, from 2019 to 2023. It separately settled charges for violating the whistleblower protection rule, by requiring departing staff to falsely attest they had never filed a complaint with a government agency. Two Sigma Investments LP and Two Sigma Advisers LP paid $90 million in penalties combined and had already voluntarily repaid $165 million to affected clients during the investigation (SEC press release 2025-15).

In September 2025 the SEC separately charged a former Two Sigma researcher, Jian Wu, with deliberately manipulating those same models between November 2021 and August 2023. The alleged goal was simple. Replicate other models’ forecasts without authorization. The SEC’s own release ties the scheme to “at least $165 million in harm to certain clients” (SEC litigation release LR-26398). Here is how I am reading these two matters together. Wu’s alleged manipulation window sits entirely inside the settlement’s broader four year vulnerability window. Both releases cite the identical $165 million figure. Same number, twice. My read is these are very likely one underlying root cause, described across two separate enforcement actions filed eight months apart, rather than two independent examples of the same failure mode. Treated that way, it is still a real data point. The facts behind a $90 million penalty were known internally as early as 2019 and were not fully public until 2025. That is a several year gap. It dwarfs the co-CEO dispute’s one day one.

What $232.8 billion actually means

Two Sigma’s headline regulatory assets under management figure, $232.8 billion as of January 31, 2026, is gross. It includes leverage. It isn’t the same thing as net investor capital. The firm’s own ADV instructions require that distinction, and press coverage routinely drops it: both Bloomberg and Hedgeweek cited Two Sigma as an “$80 billion” or “$70 billion” fund in their own coverage of this same dispute. My guess is those smaller figures track a narrower measure, most likely net assets in a specific fund family rather than the full regulatory total, though I can’t confirm the exact reconciliation from public data alone.

I flag this because it matters for reading the PERA number correctly. $100 million divided by $232.8 billion is 0.00043. I make that four hundredths of one percent of the firm’s reported book. This was never a redemption that threatened Two Sigma’s business. Not even close. I read it as a signal. It was sized to be visible to other allocators watching the same dispute, small enough that Two Sigma’s balance sheet barely registers it.

The discriminator: is nine to thirteen weeks actually slow?

Here is the honest counterargument, the one a fund operations person would raise first, and I’m not going to dress it up. Nine to thirteen weeks for an institutional LP to review a governance crisis, get investment committee sign off, and route a redemption notice through a fund’s own 60-to-90-day mechanics is not slow at all. It might be close to the fastest this kind of decision can move through a public pension’s process. Maybe this is just how slow institutions are. Under that reading, there is no real gap to explain. No gap at all. PERA may have acted as fast as its own machinery allows. The story could be unremarkable.

I take that seriously, and I cannot fully rule it out, because PERA has never stated on the record what date it filed its redemption notice, what its investment committee discussed, or when. What keeps me from fully accepting the fastest possible speed reading is the fund’s own quoted rationale: “too much uncertainty, with no clear timetable for resolving the internal conflict.” That reads as language describing what is still unresolved, not language describing a decision already made and simply awaiting paperwork. Picture PERA deciding by early April and spending the rest of the gap purely on notice mechanics. I’d expect a different sentence from them. Something closer to “the dispute remained unresolved when our notice period expired” than “no clear timetable.” That’s a real distinction. It’s PERA’s own choice of words, not mine.

A public pension the size of New Mexico’s typically runs its investment decisions through a standing investment committee. It meets on a fixed monthly or quarterly calendar. Not on demand the day a story breaks. Even a fast institutional reaction to the April news would likely need to clear at least one scheduled committee meeting. Only then could a formal redemption instruction go out. I’d guess that alone accounts for several weeks of the gap. I say guess deliberately. PERA’s specific committee calendar for this period is not something that could be verified from public sources, and naming the assumption seemed better than dressing it up as a fact.

The narrative disclosure itself is a separate, cheap check any allocator can run before a gap like this one ever opens. I’ve written up the ninety second ODD test that FINRA’s newer cyber reporting channel makes worth running on any manager, and it applies just as well to a governance narrative buried in Schedule D.

Three things I can’t rule out

Three real gaps remain. I don’t know PERA’s exact redemption date within June, or its exact investment committee timeline. Without that, “nine to thirteen weeks” is a range built from one confirmed endpoint, the April news, and one imprecise endpoint, the month of June. It is a range, not a single confirmed figure.

I don’t know Two Sigma’s specific redemption terms for the Absolute Return Enhanced Fund. The 30-to-90-day, quarterly redemption framing is an industry standard inference from Proskauer’s general market description, not a fact about this specific fund’s governing documents, which aren’t public.

One redemption isn’t a base rate. This piece shows one institutional LP’s reaction time to one piece of unambiguous, fast reported news. It doesn’t establish how quickly allocators generally move once a story like this breaks, or whether PERA was typical or unusually deliberate.

What would prove this wrong

Know your own redemption clock before you need it

Two Sigma’s Absolute Return Enhanced Fund is still up double digits this year. The $232.8 billion the firm reports managing hasn’t visibly moved because of any of this. What changed is that a $100 million allocator decided a fund’s returns were no longer the only variable worth pricing. It made that call on information that had been sitting in the trade press, cited to the firm’s own SEC filing, for weeks before anyone pulled a wire.

If you’re allocated to a multistrategy or quant manager with a founder led or dual CEO structure, the actionable version of this isn’t “watch for hidden filings.” The filing wasn’t hidden; Bloomberg had it in a day. The actionable version is narrower: know your own fund’s redemption notice period before you need it, and know how many committee meetings stand between “we should redeem” and a notice actually going out. Both numbers are already sitting in your own subscription agreement and your committee’s standing calendar, not in anything Two Sigma or PERA controls. Those two numbers, added together, are the real distance between a headline and a wire transfer, and most allocators only learn them under pressure, mid crisis, instead of in advance.

The Decision-Grade Version

This piece is complete on its own. The thesis, the evidence, and what would kill the view are all above, and nothing was held back to sell you a next step.

The Patreon note is a different piece of work, not a deeper cut of this article. It takes one method from inside this story, a five input audit for measuring your own redemption latency gap before a governance headline forces the question, and writes it the way an ODD team would actually run it: pull the manager’s own Form ADV, size your fund’s notice period, size your committee’s cadence, add them together. Written for people who sit on an investment committee or run operational due diligence for a living.

→ Read the redemption latency audit note

→ Or join the Patreon community for every note

More research like this: my YouTube channel breaks down the filings on screen, and I post shorter reads on LinkedIn.

If your manager’s governance blew up on Bloomberg tomorrow, how many weeks would your own process need before the money actually moved?

Cover photograph: F McGady, CC BY 4.0, via Wikimedia Commons.

Millennium isn’t harvesting a clean success story when it absorbs Jain Global’s outside capital. It’s converting real redemption pressure, a $250 million sovereign-fund withdrawal and a fund’s own admission that it couldn’t reach scale alone, into a deal that captures the fee economics for itself. The trading was never broken. The vehicle that housed it was.

Below the paid line:

• The redemption pressure the headline “harvest” story leaves out: GIC’s $250 million withdrawal, a January 2026 drawdown, and Jain’s own analyst-source naming redemption risk as the reason for the deal

• The corrected arithmetic separating the two real gaps in Jain’s own numbers: the ~7-point capital-timing gap and the ~33-point fee-drag gap, with the fee-waterfall caveats stated plainly

• Kodai Capital's April 2025 wind-down, a $690 million Millennium-backed manager run by Citadel alumni, and why its market-driven collapse complicates any claim of a clean pattern

• The industry pass-through comparison table: Jain, Balyasny, Citadel and ExodusPoint, each with its own disclosed gross-to-net gap

• Jain Global's tiered performance-fee schedule (10% at $250 million-plus commitments) and its 18-month lock-up plus 5% redemption-fee terms, correctly sourced

• The falsifier: what I’m watching for in the next twelve months to know if this is a repeatable playbook

I’ve covered pod shops on this desk before. This is the first time one let me see, in its own disclosed numbers, exactly how much of its investors’ money it took to build the thing before someone else decided it was worth owning outright.

Below: the 32.6-point Jain gap split into its two causes, Millennium and Citadel's own ~10% mature pass-through ratio, the four-fund comparison table, and my September 2027 falsifier date.

The consensus: platform seeding is a smart, symbiotic trade

Talk to anyone who allocates to multi-strategy funds and you’ll get roughly the same account of why Millennium, Citadel and Point72 back independent launches like Jain Global. The platform gets diversification without building a new desk from scratch. The founding portfolio manager gets a credible anchor investor and instant credibility with the rest of the capital-raising circuit.

Outside LPs get early access to a launch that would otherwise be closed to them, priced at whatever terms the manager can command in a hot market. The endowments, sovereign funds and family offices that filled out Jain’s $5.3 billion raise in 2024 are sophisticated allocators who understand pass-through fee structures going in (Business Today). Not first-time buyers.

Frame the seeding relationship as a long-term partnership, and a later decision to consolidate that manager’s book reads as a natural next step, never an extraction. A tidy story, on paper.

I think that account is broadly right about the incentives on day one. My read is that it stops being the whole story once you have two years of numbers in front of you, and I want to walk through why.

The variant view: a redemption story and a fee-harvest story, at once

Here’s the observation that complicates the clean version of this story. Jain Global wasn’t simply thriving when Millennium moved to absorb it, and it wasn’t simply failing either. Both things were true at once, and the article that only tells one half is missing the mechanism.

On the trading side: the fund generated $1.8 billion in gross trading profit against $5.3 billion raised, a cumulative gross return in the mid-thirties to low-forties depending on which capital base you measure against (Hedgeweek). That is not a broken strategy. Profitable, the whole way through.

On the LP side, the picture looks different. Singapore’s GIC, one of Jain’s earliest and largest backers, pulled $250 million in February 2026 (Hedgeweek). That same month, the fund was down 0.9%. A sovereign investor, heading for the door. Rupak Ghose, the analyst whose pass-through estimates this piece leans on, states plainly what was actually driving the Millennium conversation, in his own words:

That’s the article’s own best source, and it cuts against my own framing more than I’d like: on redemption risk, as the actual trigger, I’m not going to dress it up or write around it. I don’t think it kills the fee-harvest thesis either. The two readings don’t cancel out, and the arithmetic below is what shows why.

Jain isn’t the only recent case, either, though the pattern is messier than a clean two-for-two. About four weeks before Jain’s deal closed, in consecutive calendar months, Millennium struck a similar arrangement with Albar Capital Partners, a $1.5 billion market-neutral equities manager run by Javier Velazquez, a former Millennium portfolio manager. Albar’s deal was disclosed August 7, 2026; Jain’s was reported complete September 1-2, twenty-five days later. Albar is returning all outside capital and will run exclusively for Millennium going forward (Hedgeweek, Aug 7, 2026).

Hedgeweek’s own framing of the Albar deal names the pattern directly, unprompted: Millennium is “identifying successful portfolio managers, backing them with external capital and subsequently bringing those relationships into its own capital base” (Hedgeweek).

There’s an earlier case too, and it cuts the other way. In April 2025, Millennium-backed Kodai Capital Management, a roughly $690 million manager run by Citadel alumni Neville Shah and Joon Park, wound down and returned its non-Millennium capital amid a brutal stretch for tech names (Bloomberg). Kodai doesn’t fit “harvest of mature success” at all. A weak-market wind-down, plainly.

Laid out by date, the three read as a widening pattern, not a single template: Kodai in April 2025, driven by markets. Albar in August 2026, closer to a clean harvest. Jain in September 2026, both at once.

The honest count is three cases, not two, and they don’t all rhyme. What connects all three is simpler and less flattering than “playbook”: Millennium absorbs a seeded manager’s book whenever the manager’s outside capital becomes a liability, whether the reason is weak markets, LP anxiety, or a fee structure eating the fund alive.

The capital-timing gap: 34% versus 41%

Start with the piece of the arithmetic that is actually about timing, not fees. Real, but not the story. I want it addressed on its own, and I do not want it buried inside the bigger one that follows. Jain raised $5.3 billion in commitments, but a fund cannot deploy $5.3 billion on day one. Capital ramps in as the team hires, the risk book fills out, the operational plumbing gets built.

Hedgeweek’s own reporting draws exactly this distinction: gross return on capital raised comes to 34%, gross return on capital actually deployed comes to 41% (Hedgeweek). That seven-point gap is the cost of holding uninvested dry powder during a roughly twenty month ramp. A real drag. Nobody’s fault. What a launch this size looks like in its first two years.

I am raising this gap first because it is the one a skeptic can wave away easily, and I would rather it get waved away correctly than folded silently into the bigger number that follows. Every large multi-strat launch of the past decade has carried some version of this drag, ExodusPoint’s own 2018 ramp included. None of it tells you anything except that raising $5 billion faster than you can prudently deploy it costs early investors a few points of blended return. That part of the story is boring. It should stay that way.

The fee-drag gap: 41% versus 8.4%

This is where the story turns. Even measured on the more generous deployed-capital basis, gross return still runs 41% against a compounded net return to investors of 8.4%. I checked that 8.4% against the three annual numbers Hedgeweek reported separately: 0.5% for the six months from launch in July 2024 and 3.7% for full-year 2025 in one piece (Hedgeweek, Jan 2026), and roughly 4% year to date in 2026 in a second, later piece (Hedgeweek, Sept 2026). The January piece predates 2026 entirely and cannot support that third figure on its own.

Compound those three: 1.005 times 1.037 times 1.04. I make that 1.0839, an 8.39% cumulative net return. Close enough to the disclosed 8.4% that I trust this is a real geometric compounding of reported NAV numbers. Not a rounded talking point someone pulled from a press release.

So the gap that actually needs explaining is 41% gross against 8.4% net, 32.6 points on the precise arithmetic. Call it 33. Two separate charges sit inside it. Neither one is small.

The first is the uncapped pass-through expense structure that has become standard at the largest multi-strat platforms: trading technology, compliance, data licenses, portfolio manager compensation, office costs, billed straight to the fund instead of absorbed into a flat management fee. Rupak Ghose estimates Jain’s pass-through ratio ran as high as 15% of assets in 2025 alone, against an average deployed AUM in the $3.5 to $4 billion range that year, and projects it falling to roughly 12% in 2026 as the cost base spreads over a larger book (Rupak Ghose). Fifteen points of expense ratio in a single year, against a fund that gained mid-teens gross that same year, is close to the whole story by itself. Almost the whole gap, in one line. I can’t reconstruct the exact fee-waterfall order that turns those two figures into the reported 3.7% net (see the limits section below).

The second charge is the fund’s own performance fee, tiered by ticket size at launch: 10% for commitments of $250 million or more, dropping in steps for smaller tickets, alongside an 18-month lock-up before any redemption right opens (Hedgeweek). That same lock-up structure explains why outside LPs, GIC included, largely didn’t exit before 2026: an early withdrawal carried a further performance-fee step-up plus a separate 5% redemption fee on top of whatever the market had already taken (Hedgeweek). Staying in was the cheaper option even for an unhappy LP. A lock-up, doing quiet work. Exit was expensive too.

Jain Global has not disclosed the exact order in which pass-through costs and performance fees come out of gross trading gains, and I am not going to pretend I know it. What I can say: an expense ratio running close to gross returns in a strong year, plus a performance fee stacked on top of whatever survives it, is large enough on its own to close most of a 33-point gap. No third explanation needed.

Read more

The claim: The consensus is watching the wrong number. A basis book dies when its financing dies, never because the bond it holds fell in price, and financing is published every morning, free, at T+1.

The numbers: Leveraged money is short $823.6 billion of Treasury futures notional (CFTC Traders in Financial Futures, 25 Aug 2026, my reconstruction from contract counts). SOFR’s 99th percentile spread has held a 7 to 9 basis point band since 14 August. On 16 March 2020, the one day this trade actually broke, that spread went from 12 basis points to 174 in a single session.

The catalyst: On 31 August 2026 the CFTC pushed the Form PF amendments to July 2027, a fourth extension, while the US 30-year printed 5.25% and Japan’s 10-year crossed 3% for the first time since 1996.

Wrong if: SOFR’s 99th percentile spread closes above 20 basis points for two consecutive sessions, or Standing Repo Facility take up clears $30 billion on a non quarter end day. Both are public and free, and neither has happened.

Below the paid line:

• The full five dial monitor with every endpoint URL, the exact JSON field to read, and a threshold on each set against that dial’s own trailing distribution

• The 16 March 2020 SOFR percentile table, the signature a real unwind leaves in the funding data

• My contract by contract notional reconstruction: 7,003,041 contracts to $823.6 billion of net short, with the CME face values and the arithmetic shown

• The haircut arithmetic: what a one point widening calls in cash across the whole book, and why fifty to one is defensible until it’s not

• The four week direction of the leveraged net short, which the consensus never quotes

• The bilateral repo blind spot, and the one condition under which this entire read fails

The consensus: a record 7,003,041-contract short into the worst rout since 2007

Here’s the view I’m departing from, in the strongest form I can write it.

Global yields are repricing hard. Japan’s 10-year crossed 3% for the first time since 1996, euro area inflation printed above 3%, and the September FOMC now carries a live hike under Kevin Warsh. Against that backdrop, hedge funds hold the largest leveraged Treasury futures position ever recorded. The precedent is documented and recent: in March 2020 the cash futures basis trade unwound into a Treasury market that stopped functioning, and the Federal Reserve had to buy the market to stop it. So the size of the leveraged position is the exposure that matters, and that exposure sits at a record.

I want to be clear that I find this argument coherent, and that serious people hold it. It’s also old. Back in November 2023 Ken Griffin was already arguing that regulators worried about basis trade risk should look at bank balance sheets instead of at the funds, and the SEC’s scrutiny of the trade was public enough by then to draw that response. Three years on, the size number is the figure this market has worked hardest in this market. The consensus has the size right. My disagreement is about what size does.

The variant view: the position is quarterly and stale, the financing is daily and free

My claim is narrow and I’d rather state it narrowly than oversell it.

A leveraged Treasury book doesn’t liquidate because the asset it holds fell in price. It liquidates when somebody stops lending against that asset, or lends at a haircut it can no longer meet. Price and financing are different variables. They correlate in a crisis, which is exactly why people conflate them, and that conflation is the whole mistake.

Here’s what turns my disagreement into an edge instead of a debating point. The position is observable quarterly, two quarters late. The financing is observable daily, one day late, at no cost. Everybody is watching the slow one. It gets worse for the slow one. On 31 August 2026 the CFTC and the SEC published a joint final rule (PR 9290-26) moving the compliance date for the Form PF amendments from 1 October 2026 to 1 July 2027. Read that carefully, because the detail is the story: those amendments were adopted on 8 February 2024. They are not a proposal anybody is still arguing about. They are settled law whose start date has now been pushed out again, and the trade press counts this as the fourth such push. The OFR’s Hedge Fund Monitor, which is built on Form PF, currently runs to 31 March 2026. Five months stale. And the upgrade everyone has been waiting two and a half years for just slid another nine months.

So the profession is underwriting a number that’s five months old and won’t improve until 2027, while ignoring a number that refreshed at eight o’clock this morning. I think that asymmetry is the trade.

The mechanism: three ways a basis book actually dies

You know the trade: long the cash bond, short the future, earn the spread, lever it through repo because the spread is thin. Fifty to one is ordinary, and on its own terms defensible, because the spread carries very little risk if you can hold to convergence. If you can hold. That conditional is the entire risk, and it is the part the coverage skips.

Three things stop the position being holdable. None of them is “the bond fell.”

One: the repo lender raises the haircut. The fund borrowed 99 cents against a dollar of collateral. The lender now wants to lend 97. Two cents per dollar, across the whole book, in cash, today. Nothing about the bond’s yield caused that.

Two: the futures leg calls variation margin faster than the cash leg pays. The legs offset economically. They do not offset operationally. The short future is marked daily through the clearing house and settles in cash, while the long bond gains value on paper. A fund can sit perfectly hedged and still run out of money on a Tuesday.

Three: repo stops clearing at a single price. This one matters most and gets skipped most. Repo is a distribution, and the tail of that distribution is where a borrower who has lost the market’s confidence actually funds. Somebody being forced out pays up in the tail well before the average moves. So the tail moves first, and it moves violently. That third channel is why I watch a percentile instead of a rate.

Sizing the book from contract counts, because nobody hands you the number

The consensus quotes 7,003,041 contracts as though a contract count were a position. It is not one. Converting it is worth doing yourself, because the conversion is where the risk actually lives.

CME Treasury futures carry fixed face values. The 2-year note contract is $200,000 of face. The 5-year, 10-year, Ultra 10-year, classic bond and Ultra bond contracts are $100,000 apiece. I multiplied the leveraged money positions from 25 August by those face values:

Gross short notional of $1,070.9 billion against gross long of $247.3 billion, for a net short of $823.6 billion. Before that number does any work in your head, take the concession that governs it: the CFTC’s “leveraged money” bucket holds CTAs, macro funds and anybody trading with leverage, so this is an upper bound on basis positioning and not a measurement of it. Some unknown share of that short is trend followers who have never touched a cash bond. I use it anyway because its DIRECTION is informative even when its composition is not, and I flag below where that reasoning could fail. Now the second caveat. This is the FUTURES leg only, reconstructed from contract counts and published face values, with no proprietary feed anywhere in the chain. It is not an estimate of the basis trade, and I want to be exact about that, because published estimates of the trade itself range widely by method and vintage. Work built on OFR data put the cash futures basis near $830 billion as of September 2025, a figure I walked through in detail when Citadel’s own SEC filing came in at $570 billion against the $71 billion everyone prints; press estimates circulating in January 2026 ran closer to $1.5 trillion. Those are different measures of different things, and I have not reconciled them here. What my $823.6 billion gives you is something none of them do: a number you can rebuild yourself, from one free file, every Friday at 15:30. Reproducibility is the property I’m buying here, not precision.

Now put the haircut mechanic against that number, because this is where leverage becomes arithmetic instead of an adjective. Take a clean $1 billion of cash Treasuries financed in repo at a 2% haircut. The fund posts $20 million of equity and borrows $980 million. Fifty to one. Move the haircut to 3%. Required equity goes to $30 million, so the fund owes $10 million per billion, in cash, that day. The bond did not change. No yield moved. A lender revised a percentage. Scale that across $823.6 billion and a one point haircut widening calls $8.2 billion of cash into existence overnight. That is the mechanism in full, and it is why I keep insisting the financing dial outranks the price dial. The haircut is the trigger. The yield is weather.

One more link in the chain makes it monitorable. A lender who is widening haircuts is also repricing in repo, and repo reprices in the tail before the average. Which brings me back to a percentile.

Read more

The claim: The number every outlet ran this week, $250,000 into $60 million, a 240x gross return, requires Thirty Five Ventures to hold 0.465% of Hugging Face at closing. No filing, no company statement and no spokesperson has disclosed a percentage. The two documents Hugging Face did file with the SEC, plus its own funding announcements, put Kevin Durant in four financing rounds. The multiple everyone quotes uses two.

The numbers: Hugging Face’s 2018 seed Form D reports $3,315,000 sold against a $4,000,000 offering, to six investors, in debt. The 2019 Series A Form D reports $19,710,523 sold to 30 investors and states on its face that the figure includes conversion of the earlier notes. $60,000,000 divided by $12,900,000,000 is 0.465%.

The catalyst: The acquisition is reported and unconfirmed, and the two outlets that broke it disagree. The Information reports an agreed deal at $12.9bn; Business Insider reports talks that “had not yet produced a signed agreement.” The payout stories ran on the 27th and 28th.

Wrong if: Thirty Five Ventures discloses that its Series B and Series C checks were immaterial, or Durant’s closing stake is confirmed near 0.465% on a $250,000 basis. Either kills this and I withdraw it.

Nvidia has agreed to buy Hugging Face, the repository where open AI models are hosted and downloaded, for $12.9 billion. Inside a day the second story had eaten the first. Kevin Durant put in $250,000. Kevin Durant is getting $60 million. Twenty-four thousand percent.

I went looking for the cap table behind that arithmetic, expecting a disclosed stake and a lazy multiplication on top. The stake isn’t there. So I pulled Hugging Face’s own SEC filings and read the fields one at a time, then checked the investor list of every round it has announced. Both contradict the coverage.

Below the paid line:

• The $60,000,000 sensitivity table line by line: what 240x becomes on a $500,000 basis then a $750,000 basis then $1,250,000 and finally $2,250,000 against the 6.45x ceiling every Series C dollar drags it toward

• Both Form Ds field by field: the isDebtType flag that makes the seed’s ownership unobservable, and the clarification that counts $3,315,000 twice

• The verbatim investor lists from TechCrunch’s March 2021 Series B and Hugging Face’s own Series C post, the two documents that break the denominator

• Why both headline numbers are wrong and move in opposite directions, plus the 0.465% stake the payout silently requires

• Nvidia’s own position in the $235 million Series D, and the $42.3 billion of unnamed private stakes on its balance sheet

• Four confounds I cannot rule out, including the 10% to 20% escrow convention and the unsigned agreement

The $60 million traces to four words, and neither of them is a filing

Provenance first, because everything else depends on it. Bleacher Report attributes the figures to one origin: “Insider Joe Pompliano reported Thursday that Durant and his business partner, Rich Kleiman, invested $100,000 in the seed round and $150,000 in the Series A round.” The payout sentence is quoted from him: “He likely made more than $60 million in this investment alone.”

Two words carry the whole edifice, as I read it. I’m told, for the check sizes. Likely, for the outcome. That’s the entire evidentiary base, and “one guy on X” is not automatically wrong: Pompliano covers this beat and may have a real source.

I chased the attribution and it doesn’t converge. Bleacher Report credits Pompliano. TheNextWeb credits Front Office Sports and says only “early rounds.” Yahoo Sports credits Front Office Sports for the $250,000, then names nobody for the payout, which it reports as “believed to be upwards of $60 million.” Three chains, none terminating in a document, two of them sharing an origin. No SEC filing anywhere, no Hugging Face statement, no Nvidia statement, no comment from Durant or Kleiman.

Then I watched the hedges fall off. Joe Pompliano wrote “I’m told” and “likely made more than $60 million.” Anthony Pompliano, to a much larger audience, posted “Kevin Durant made $60 million from investing $250,000 in Hugging Face. Insane.” No hedge, no conditional. The whole mechanism in two posts.

Hugging Face filed $3,315,000, and it was a note

Hugging Face has filed exactly two documents with the SEC in its life. EDGAR returns “Items 1 - 2” for CIK 0001739138, both Form D. The last is dated 12 December 2019.

I pulled the 2018 filing. An offering of $4,000,000, a total amount sold of $3,315,000, first sale 20 April 2018, investors already invested: six. Signed by Julien Chaumond as Treasurer, out of 81 Prospect Street in Brooklyn. Clément Delangue’s name is misspelled “Delange” in the related persons block, which tells you a founder filled it in.

I keep reading “a $4 million seed round,” and I want to be careful about what the filing does and doesn’t prove. Per the SEC’s Form D FAQ, a later change in “the amount of securities sold in the offering” is among the changes that do not require an amendment. The round may well have closed at the full $4 million with nobody obliged to say so. The press figure is the target. The filed figure is the only dated, sourced number in the story.

Here’s the larger one. Under types of securities offered, the 2018 filing marks isDebtType true and leaves the equity flag unset. The seed money bought a convertible instrument. A note or a SAFE, and no stock.

That one field changes what I can say a $100,000 check means. Equity at a price buys a known percentage the day it clears. A note buys a claim that converts later, on terms set by a valuation cap and a discount, and neither appears in a Form D. Nobody outside the cap table can compute what that $100,000 became. That’s a property of the instrument.

The Series A Form D counts the seed money a second time

The 2019 Form D reports $19,710,523 sold to 30 investors, first sale 27 November 2019, equity this time. TechCrunch reported it as a “$15 million funding round” led by Lux Capital, with “A.Capital, Betaworks, Richard Socher, Greg Brockman, Kevin Durant and others.”

The filing explains its own gap, in a clarification field I doubt many people open:

Offering Amount and Amount Sold includes the conversion of convertible notes and Simple Agreements for Future Equity previously sold by the Company into the Equity issued at this offering.

So the $19.71 million isn’t $19.71 million of new money. It holds the 2018 round arriving a second time. Net the $3,315,000 of converting seed paper against it and you get roughly $16.4 million of fresh equity, close enough to the reported $15 million that the two accounts reconcile. Sum the Form D amounts to get “total raised” and you overstate Hugging Face by $3.3 million.

There’s a subtler consequence. The conversion happened at the Series A, so the note’s cap was applied on 27 November 2019 and the share count folded into that round. Durant’s seed and Series A positions were set in one event, and I can’t find their combined percentage published anywhere.

$60 million back solves to 0.465%, and nobody has disclosed a percentage

Now do the division I could not find anywhere in the coverage.

$60,000,000 divided by $12,900,000,000 equals 0.004651. Call it 0.465% of the company at closing, before any escrow, holdback or consideration mix. I give it three figures because the input has about one: every source hedges it as “likely more than” or “believed to be upwards of,” and a quotient cannot be more precise than what went into it.

That is what the reported payout requires. Requires of whom is the next problem. Bleacher Report attributes the $250,000 to “Durant and his business partner, Rich Kleiman,” others attribute it to Thirty Five Ventures, and all of them attribute the $60 million to Durant alone. The firm was founded by both men and I can find no disclosure of how its position splits from Durant’s own economics. Numerator and denominator may not describe the same holder.

Now ask what public evidence supports it. None does. Hugging Face has filed nothing since 2019, and as a private Delaware corporation it owes no 10-K, no proxy, no beneficial ownership table.

Nvidia’s own disclosure doesn’t help, and I went and read it. Note 6 of its April 2026 10-Q reports every private company stake it owns as one aggregate roll forward and names not one of them. Nvidia’s own Hugging Face stake is invisible in Nvidia’s own filings.

So I read 0.465% as a back solve wearing a measurement’s clothes. Pompliano didn’t measure a stake and multiply. He estimated an outcome, and the percentage is what falls out when you divide. I’m not saying the number is too high. My claim is narrower: nobody outside the cap table can produce it, and the coverage presents it as though somebody counted.

Read more

Zero coupon convertible issuance is being reported as free financing that AI-era stock volatility has made possible. The primary filings show issuers paying real, quantifiable cash for it, and the single deal every tracker cites as proof of the record year carries a 6.25% coupon.

Below the paid line:
- The capped call cost table across Penguin Solutions, Alphabet’s Series B tranche, Tempus, and Bitdeer, with cap prices, premiums, and annualized cost
- The 51-filing EDGAR cross section of 2026 capped call issuers outside AI, and why that sector spread changes the story
- The mechanism proof, line by line, for why a capped call’s price reveals the value of the option the “0% coupon” headline treats as costless
- The convertible arbitrage buy side: delta hedging, gamma capture, and why dedicated arb allocation shrank from roughly 20% of a hedge fund book in 2000 to 1% to 2% today, even as issuance sets records
- The exact falsifier threshold: under 2% of face kills this thesis, and where the four name sample sits against it today

The story every desk is telling: $161.5 billion and a free option

Every desk covering convertibles this year is telling a version of the same story. I do not think it is a dumb one. Rates have stayed high through 2026. Straight high yield debt is expensive. AI-linked equities swing hard enough that the option to convert a bond into stock is worth something real to a buyer. Volatility is the product.

Numerix counted $84.6 billion across 96 global convertible deals in Q2 2026 alone. That is a 67% jump over Q2 2025. A record quarter. Of those 96 deals, 44 priced at a 0% coupon, 46%. ION Analytics puts the year to date global total at $161.5 billion through July 22, already 23.6% above the prior full year pace record set in 2021. Man Group’s own credit desk puts full year 2026 issuance on pace to exceed 2025’s record of $167 billion. The pace is real.

The mechanism analysts cite is standard convertible bond pricing theory, correct as far as it goes. A convertible’s value splits into a straight bond component and an embedded call option on the stock, and that option’s price rises with implied volatility. Push implied vol high enough and a company can sell the option for a 0% coupon while investors still buy the bond at par, paid entirely in optionality instead of cash. Barclays and BofA research, cited across the trade press, frame it as a substitution: companies that would have paid 6% to 8% on high yield debt are instead paying nothing, funded by volatility itself.

The sector concentration behind those totals is worth sitting with, because it is more lopsided than “AI is driving issuance” implies. Computers and electronics accounted for $86.23 billion of the $161.5 billion global year to date total, 53.4% of every dollar raised in the convertible market this year, across 139 deals, per ION Analytics. Healthcare ran a distant second at $12.7 billion, 7.9%. Not close. AI-adjacent supply is not one flavor among several here. It is close to being the whole market for 2026, which is exactly why a mispriced assumption inside that sector’s dominant instrument would matter more this year than in almost any prior one.

My read is that this framing survives exactly as long as nobody opens the prospectus supplement.

Alphabet’s own straight debt told a related story earlier this year: a $20 billion bond offering priced spread compression tight, without any embedded option at all.

Penguin Solutions spent $49.1 million to buy back what it just sold for nothing

I went looking for the cost of that “free” option, because a financial instrument does not cost nothing just because the number on its face is zero. What I found in the filings is that issuers are not treating their own zero coupon deals as free. They are spending real cash, disclosed in dollars, to buy a large chunk of that option straight back.

Penguin Solutions issued $750.0 million of 0.00% Convertible Senior Notes due 2031 on July 17, 2026. The notes carry no regular interest and do not accrete principal, about as close to a literal zero cost instrument as debt gets on paper. Four paragraphs later, in the same 8-K, the company discloses it spent approximately $49.1 million, 6.55% of face value, on capped call transactions to reduce the dilution those same notes create. That is not a rounding error. It is the company underwriting the accuracy of the “free money” pitch with its own cash, and I do not think it fully believes the pitch.

Alphabet’s Series B Mandatory Convertible Preferred Stock, priced the same quarter at a real 6.25% annual dividend, still carried $439.7 million of capped call spending on the base tranche alone, per its own 424B5. Against that tranche’s roughly $8.375 billion in gross proceeds, that is 5.25%. Bitdeer Technologies, the crypto mining company that has spent the past two years converting mining capacity into AI and HPC cloud infrastructure, spent $29.2 million on capped calls against $315.1 million of net proceeds from a February 2026 convertible refinancing. That is 9.27%, the highest ratio of the four names here, on notes that carry a real coupon. Tempus AI, the healthcare data company, priced a true 0.00% coupon note on May 7, 2026 and spent $27.2 million against $384.1 million in net proceeds on the base offering, 7.08%.

Four unrelated issuers. Four separate deals, four different coupon structures. Four capped call costs, all clustered between 5% and 9% of proceeds. That clustering, across issuers who share almost nothing else, is the tell. Four strangers, one number.

The capped call is a receipt: what real dollars reveal about a “free” option

The size of the check an issuer writes for a capped call is the whole argument, so start with what the instrument actually does. Selling a convertible note makes a company short a call option on its own stock: the right the bondholder has to convert debt into equity at the conversion price. If the stock rockets, that option transfers real value from existing shareholders to bondholders, dollar for dollar, on every share issued above the conversion price. A capped call is the company buying that option back from a bank, up to a cap price, recapturing the dilution between the conversion price and the cap.

Penguin Solutions set its conversion price at $116.70 a share and its cap at $175.05, a 125% premium over its July 14 closing price of $77.80. Between $116.70 and $175.05, Penguin is now hedged. Insurance costs money. Every time. Above $175.05, it is exposed again. The price of that hedge was not a fee for paperwork. $49.1 million is the market’s own valuation of the equity optionality Penguin just wrote into its notes, priced by the banks who structured the deal. A dealer’s own margin sits inside that price too, so I read it as a floor on the option’s value rather than a clean clearing price. The floor still runs at real money.

Here is the arithmetic the “0% coupon” headline erases. Amortize Penguin’s $49.1 million over the note’s five year life to maturity in 2031, and you get roughly $9.8 million a year. That is 1.31% of face value annually. Not the highest cost of capital in the market. Also not zero. The company’s own transaction structure proves it.

I make Alphabet’s Series B capped call spend, $439.7 million against roughly $8.375 billion of base tranche gross proceeds, at 5.25%. That sits close to Penguin’s 6.55%, even though Alphabet is investment grade and pays a real coupon that Penguin’s notes carry none of. Two issuers, wildly different credit quality, land within 130 basis points of each other. Credit rating barely moved the price. Vol did.

The reason: a capped call’s cost tracks implied volatility on the underlying stock far more than it tracks the issuer’s credit rating. That’s precisely the variable AI-adjacent equity has pushed to multi year highs, across mega caps and mid caps alike. Bitdeer’s higher ratio, 9.27%, fits the same logic from the other side. A smaller, more volatile name pays more per unit of dilution protection than Alphabet does. Coupon structure doesn’t enter into it. Vol sets the price.

Read more

The claim: Venezuela’s reported 100-year oil lease is being priced by the market as a durable commitment built to outlast any one government. I’d price it as the opposite signal. At any discount rate a Venezuelan sovereign risk deserves, the back eighty years of a 100-year lease are worth almost nothing today.

Consensus: a retirement party, not a departure

The reporting on Venezuela’s possible OPEC exit has converged on one reading, and it is a reasonable one. Bloomberg’s own account, carried by Investing.com, notes the country’s departure “would have limited immediate impact on global oil markets,” given how little it produces today against its own past output. Venezuela has sat outside OPEC’s formal quota compliance since 2016, when collapsing output made its assigned target meaningless. It has produced without reference to a cartel ceiling ever since. A country that already ignores its quota loses nothing by making the exemption official. Nothing lost, nothing new.

The UAE left OPEC+ in May for capacity reasons of its own. It was producing 3.4 million b/d against an estimated 4.2 million of effective capacity, and it wants 5 million by 2027, which the quota it just left would never have allowed. That exit barely moved the strip. Venezuela’s, on this reading, is a smaller version of the same story: paperwork catching up to a relationship that’s already public, one confirmed departure and one live deliberation raising a legitimate but slower moving question about how much pricing power Saudi Arabia and Russia have left. The sell-side version of this same smaller story is narrower and fair on its own terms: Citi’s Scott Gruber told Bloomberg the oilfield services opportunity is real regardless of who owns the lease, a services revenue argument, not a reserve ownership one, and a drilling contractor paid this quarter does not carry the same risk as an equity holder in year 41. A different risk entirely.

There’s a second, stronger version of the consensus, and it deserves a real hearing rather than a dismissal. This time, the argument goes, actually is different. Nicolás Maduro was captured in a US raid on January 3, and Washington has controlled the marketing of Venezuelan crude since, with President Trump saying in late July that sales had already cleared $13 billion. Energy Secretary Chris Wright put the logic plainly at a Goldman Sachs energy conference: the US intends to control Venezuelan oil sales “indefinitely,” saying it needs that leverage to drive change in Venezuela. A reform of the Hydrocarbons Law took effect January 29, letting private companies hold majority upstream positions for the first time since the 2007 mixed company structure. Chevron’s Venezuela output is already up 15% over six months, guided toward 50% growth by the end of 2028. The growth is genuine.

On the strongest telling, a 100-year lease is not bravado. It is the natural conclusion of a government that finally has both the legal framework and the foreign backing to commit for the long run, and the earlier reversals belong to a Venezuela that no longer exists. I take that case seriously. But it is not the discriminating question, in my view. The discriminating question is what happens once US attention moves on, and nothing about a 100-year number answers it.

The variant view: the number is the tell

Here’s what the “no impact” framing misses. Venezuela’s exit changes nothing about barrels this year. That’s true. It is not evidence that nothing about the exit matters. Small does not mean irrelevant.

The 17-field lease being negotiated alongside it is the actual story. Its headline term, a century, isn’t a sign of confidence. It’s a sign the number has stopped doing economic work and started doing political work. No oil company prices cash flow a hundred years out. My read of the reserve math below shows why: at any discount rate that belongs on Venezuelan risk, years 41 through 100 of a lease are worth low single digits of its total value. Almost nothing, by design.

A company that actually expected to collect for a century wouldn’t need to say so. The contract would just run its economically relevant course and get renewed, the way Chevron’s existing Venezuela agreements already do: terms expiring in 2039, 2041, 2047 and 2050, all comfortably inside the window where the arithmetic still matters. A hundred years isn’t a cash flow assumption. It’s a claim about permanence, aimed at persuading investors, lenders, and Venezuela’s own public that this arrangement won’t be reversed the way the last two were. That claim would only need making if the market’s actual prior is that it will be. A tell, not a guarantee.

That judgment already exists, and it’s the sharpest data point in this whole story, though not for the reason it first looks like. Chevron’s own 10-K for fiscal 2024 states it without qualification: “As of December 31, 2024, no proved reserves are recognized for these interests” in Venezuela. This from the one company that never left, that operated straight through Chávez’s 2007 nationalization. It carried roughly 160 million barrels of proved Venezuela reserves on its books right up until 2020, when it took a $2.6 billion full impairment and removed every one of them. Gone, all at once.

Here’s the part I think actually matters: what triggered that impairment wasn’t a Venezuelan expropriation. It was Washington. The US sanctioned PDVSA, Chevron’s own joint venture partner, in January 2019, and months of narrowing licenses later, Chevron’s accountants concluded the loss was no longer temporary. The reserve line has never been restored since, not through five years of licenses granted, revoked and regranted, most recently when the White House pulled Chevron’s operating license outright in February 2025 and then handed back a new one months later, the same relief this piece already credits with Chevron’s current output growth.

That’s the market’s actual, central view of Venezuelan title security, and it isn’t really about Venezuela at all. It is a live demonstration that Chevron’s entire Venezuela exposure runs on a license Washington can revoke on a phone call, which is exactly the discretionary account risk this piece keeps coming back to on the money side too. The reserve line was filed with the SEC under penalty of law in February 2025, months before anyone was discussing a 100-year lease, and it still has not moved. The lease term is a marketing number. I’d treat the reserve line as the real one, for what it actually measures. The filing is the fact.

The mechanism: why 100 years is not worth what it sounds like

Start with the arithmetic, because it is the whole case. A lease’s value is the sum of its future cash flows discounted back to today, and the discount rate does almost all the work in a story this long. Venezuelan sovereign risk, even under the current arrangement, sits in the 8-to-12% range a frontier energy asset typically carries, in my judgment. Nothing exotic about that number.

I ran the standard annuity formula, present value of $1 a year for N years at rate r equals one minus (1+r) to the power of negative N, divided by r, across a full 100-year term against shorter windows. The split is stark. At a 10% discount rate, the first 40 years capture 97.8% of a 100-year lease’s entire value. The last 60 years add just 2.2%. Push the rate to 12%, appropriate for a deal this exposed politically, and years 41 through 100 shrink to 1.1% of total value. Even at a generous 8%, they are only 4.6%. The tail barely moves the total.

I make that the back six decades of “a hundred years” worth somewhere between one and five cents on every dollar the headline implies.

I walk through the full worked example at three discount rates, plus how to apply the same framework to any long-duration sovereign concession, in The Sovereign Concession Discount.

Venezuela’s own base rate is faster than the discount curve

Now replace 40 years with Venezuela’s own measured history, because that is the number that should actually discipline this deal.

The country nationalized its oil industry outright on January 1, 1976, forming PDVSA under President Carlos Andrés Pérez. It reopened to foreign capital through the Apertura Petrolera in the early 1990s, producing the four Orinoco strategic association agreements that became Chevron’s current joint ventures. In 2007, roughly three decades after the original nationalization, Hugo Chávez converted those associations into PDVSA-majority “mixed companies,” and ExxonMobil and ConocoPhillips walked rather than accept the new terms. Chevron, Statoil, BP and Total stayed. Nineteen years after that reversal, the Hydrocarbons Law reform reopened the sector again, this January. The cycle turns again.

Three cycles. One direction each time. The three gaps, 1976 to the early-1990s opening, that opening to 2007, and 2007 to this January, each run somewhere between fourteen and nineteen years. I will use seventeen as the representative gap for the arithmetic below, the middle of that range. Seventeen years, not a hundred.

At a 10% discount rate, the first seventeen years of a 100-year lease already capture 80.2% of its total value. Whatever Venezuela’s own history says about how long an opening actually lasts, I would argue the market has already been paid most of what the full century was ever going to be worth, in present value terms, by the point that pattern would typically turn again.

None of this is unique to Venezuelan crude, as far as I can tell. It is the same math every long duration concession runs into, and the historical record backs the arithmetic instead of complicating it. William Knox D’Arcy’s 1901 concession in Persia ran 60 years on paper; Britain lost it in 1951, at year 50, when Iran nationalized the Anglo-Iranian Oil Company. Kuwait Oil Company, founded in 1934 on a concession commonly put at 75 years, was fully nationalized by the state in 1975, at year 41. Both companies held, on paper, some of the strongest legal protection a foreign concessionaire could get in their era. Both lost control decades before their contractual term ran out: D’Arcy’s grant had ten years left on paper, Kuwait’s had thirty four. Neither one made it.

No concession this long has ever actually run its full course, in either of these two cases, and I found nothing in the historical record that runs the other way. Neither one was actually a hundred years, which makes the pattern more damning for a 100-year claim, not less: even 60 and 75-year grants, shorter and on paper safer, got cut short by decades. Venezuela’s own average across its three cycles, seventeen years, is not an outlier against that record, in my read of it. It is just a faster version of the same pattern, playing out on a shorter clock. A faster clock, same story.

The discriminator: what actually makes this different, and why it cuts the other way

The strongest version of “this time is different” rests on US involvement being categorically new. Bigger, and different in kind. That’s correct as description. Where it breaks down is the assumption that direct US control makes the deal more durable. I think it does the opposite. Legitimacy at home is what actually decided the last two reversals, and legal drafting had almost nothing to do with either one. Politics decided it, not paperwork.

The 1976 and 2007 nationalizations weren’t imposed on a reluctant Venezuela. They were popular. Pérez nationalized in 1976 with broad domestic support during a high oil price boom. Chávez’s 2007 move was a central plank of a presidency that had just won reelection. Both leaders drew domestic legitimacy from taking Venezuelan oil back from foreign companies. Popular moves, both times.

The current arrangement inverts that structure completely. It follows a foreign military operation that captured a sitting head of state, installed an interim government whose own constitutional 90-day mandate lapsed in early April with no confirmed extension vote from the National Assembly, and routes oil sale proceeds through accounts the US Energy Department’s own fact sheet describes as disbursed “at the discretion of the U.S. government.” A State Department official told Congress in April that only about $3 billion of the $13 billion collected has been accounted for, and a congressional watchdog has since opened a formal review. I don’t read any of that as a stronger foundation for a hundred years of commitment than 1976 or 2007 had. I read it as weaker, because it carries no domestic electoral mandate at all. Only a foreign one. And every reversal in this piece’s history came from a government asserting that the deal in front of it wasn’t its own. No local mandate this time.

I should name the strongest counter directly: raw external power has sustained illegitimate arrangements for decades with no local consent at all. The US has held Guantánamo Bay on a lease Cuba has refused to cash the rent checks for since 1959, sixty seven years and counting, on the patron’s staying power alone. But Guantánamo is a fenced enclave needing no functioning courts, no cooperative workforce, no administrative continuity across 28 million people. A national oil sector needs all three, so I think legitimacy matters more here, but I won’t pretend the comparison is airtight. An imperfect parallel, still useful.

What the market’s own burned players think

The clearest evidence the market already agrees sits with the two companies who lived the last reversal directly. ExxonMobil and ConocoPhillips are negotiating a potential return to the 17-field opportunity, and both are visibly reluctant.

ExxonMobil chief executive Darren Woods has said the company has “had our assets seized twice,” and that returning “would require some pretty significant changes from what we’ve seen historically here,” calling Venezuelan oil “uninvestable” without “durable investment protections” and legal change.

I read that as the closest thing available to an expert market view from inside the trade. It’s priced not in a research note but in a chief executive’s own reluctance to commit shareholder capital.

I went into the arbitration record expecting two clean wins for the companies burned last time. It isn’t there. ConocoPhillips is still owed the bulk of an $8.7 billion ICSID award from the 2007 seizure, an amount that has grown past $11 billion with interest after Venezuela lost its annulment appeal. ExxonMobil’s own 2014 ICSID award started at $1.6 billion and went the other way: an annulment committee struck roughly $1.4 billion of it in 2017, leaving the company with closer to $200 million from that proceeding. Binding international arbitration was already the strong legal protection tried once, after the last cycle turned. Where it held up, collecting took the better part of a decade. Where it didn’t, the loss came from the same system that was supposed to be the guarantee. My read: a fresh 100-year lease document doesn’t erase that history. It sits directly on top of it, and the two firms with the most reason to trust a new promise are negotiating hardest against trusting it.

The limits: where the case actually weakens

I won’t pretend the discount I’m describing is total. Chevron’s growth this year is real. 280,000 b/d today, up 15% in six months, against a national total of 1.16 million b/d, means Chevron alone already accounts for close to a quarter of everything Venezuela produces. A real and growing share. Sanctions relief and the new Hydrocarbons Law are functioning changes to the operating environment. I’d count them as more than announcements.

And the degree of direct US financial control here, proceeds settling into US-controlled accounts instead of PDVSA’s own, has no precedent in any of the historical cases I’m comparing it against, whether that’s Iran under the D’Arcy concession, Kuwait under its 1934 grant, or Venezuela’s own two prior openings. If Washington’s involvement persists for years, well beyond one administration’s term, the risk profile could shift in ways my base rate doesn’t capture. A genuine unknown, stated plainly. I have no comparable case of a producing country’s oil sector run this directly by a foreign government for an extended stretch. I can’t rule out that this is a real structural break, not a repeat.

A third limit: Brent and WTI both drifted lower the same week the OPEC-exit report landed, mostly on broader demand worries, and I can’t point to desk commentary pricing a distinct Venezuela premium into the curve. I’m reading that as complacency. A fair critic would read the same silence as a reported number with no signature simply not deserving a reaction yet, and I can’t fully rebut that until a real term sheet exists.

This mispricing has a short shelf life, which I want to say plainly rather than let it read as a standing inefficiency. The moment a term sheet leaks or gets confirmed, whatever gap exists between the headline and the deal’s real economics closes in a single trading session. Fast, once it happens. That is a matter of months, in my estimate, not years, given how actively both governments are negotiating.

The other honest gap: nobody outside the talks has actually seen one. “100 years” comes from Bloomberg’s own unnamed sources alone, per reporting out of Rio Times tracking the same day’s coverage: Reuters and Axios covered the same story without citing any lease length at all, and no government has confirmed a figure. That same Rio Times piece adds the detail that does the most work for my own argument: a hundred years is, by its own account, “a length Venezuelan law does not provide.” If that’s right, the number was never going to be signed as written in the first place. Dead on arrival, maybe. Hydrocarbons minister Paula Henao, appointed in March, has publicly invited American firms to invest, but the specific fields and the specific term she has in mind haven’t been made public. The number could well get replaced by something closer to the 25-to-40-year range Chevron’s existing joint ventures already use. If that happens, the “headline theater” reading I’m making gets confirmed, not refuted. But I should say plainly, here, that I’m reacting to a reported number, not a signed one, and to one outlet’s sourcing at that.

What would change this view

Three things would move me off this read, and all three are dated and checkable, which is the point of writing them down now.

That would be real progress on the first. The second confirms my read directly; no dodging that outcome if it doesn’t hold. That third one is the real test, in my view. Every reversal in this piece’s history came from a government asserting a predecessor’s deal wasn’t its own to honor. The current arrangement hasn’t faced that test yet. It hasn’t had an elected successor yet, by construction. The test has not happened yet.

The close: price the book Chevron actually keeps

If you’re underwriting Venezuela exposure through Chevron, or through any other name that ends up inside the 17-field talks, the instrument to price isn’t the 100-year headline. It’s the paper Chevron already holds.

Joint ventures running to 2039, 2041, 2047 and 2050 sit inside a 13-to-24-year window, almost entirely within the part of the discount curve where the arithmetic still counts. A company with every commercial incentive to book Venezuela at a positive number still won’t put one on its balance sheet, because the license under it can be pulled faster than any Venezuelan government could move. Treat anything past that window as pure optionality, priced the way Chevron itself is already pricing it. At zero. I’d trust that book over this week’s headline.

So which would you rather hold: Chevron’s proven, dated paper through 2050, carried on its own books at zero, or a fresh 100-year lease priced by a negotiating table that has shown no such discipline at all?

I post this same research on YouTube and LinkedIn as it publishes.

The Decision-Grade Version

This piece is complete on its own. The thesis, the arithmetic, and what would change my mind are all above, and nothing was held back to sell you a next step.

The Patreon note is a separate piece of work rather than a deeper cut of this article. It walks the same mechanism as a general framework for pricing any long-duration sovereign concession, with the full three-rate worked example and a section on where the framework breaks that this article doesn’t have room for. Written for people who apply this kind of arithmetic to positions, not just to one headline.

→ Read the sovereign concession discount note

→ Or join the Patreon community for every note

If you strip the country name off this, is a hundred-year contract ever actually worth pricing past year forty, or is a number that long always doing more marketing than arithmetic?

Cover photograph: Dreamyshade, CC BY-SA 4.0, via Wikimedia Commons.

Below the paid line:
- The full disclosure inventory: each of the three subpoena asks mapped to the rule, form or system that already carries it, with the citation
- The 13F growth series read off five EDGAR cover pages, $2.12 billion to $20.24 billion, and the General Instruction 4 note that let Form ADV Items 5 and 7 stay lawfully wrong
- The proposed $10 billion Form PF threshold worked against this fund’s own $9,278,344,000, and the $721,656,000 that separates them
- The Archegos theory stated at full strength, in the SEC’s own charging language, and what survives it
- Rule 13h-1(d)(2)(xii) and the next morning route to trade timing that needs no subpoena
- Five confounds I can’t rule out, including the one that most damages this piece

The consensus, stated at its strongest

Here is the reading in circulation this week, in the form its own believers would sign. A fund built a concentrated, heavily financed book in AI infrastructure names and ran it through four of the largest prime brokers on the Street. In late July the trade went against it, margin calls landed, and the portfolio was sold into a market with no natural bid. Someone in that sequence may have done something wrong, and the SEC has gone to the lenders to find out.

That’s a coherent story and parts of it are certainly true. Reporting on the asks is specific. The subpoenas sought information on when Situational Awareness executed its trades and on how the fund communicated with its lenders about borrowed capital, and the banks were told to preserve records (Quartz, via Yahoo Finance). I read preservation notices as what a regulator sends when it wants the option to look later. Nobody has been accused of anything, the SEC declined to comment, and none of the four banks commented (CNBC). Both accounts are relays and it is worth knowing whose: CNBC is carrying Reuters, the Quartz piece is carrying the New York Times.

Situational Awareness’s own response fits the frame. “It is to be expected that regulators would closely examine any funds that are high profile, produce significant returns or have particularly dramatic drawdowns,” a spokesman said. “We are a highly regulated business and will cooperate to the fullest extent with any regulatory request.”

I don’t think that reading is foolish. I think it’s incomplete, and the thing that completes it never appears in the reporting. It sits in the filings.

What the filings say, and the part nobody checked

Quartz, summarising the New York Times account, carries a clause that is easy to skim past: the banks are known “according to a regulatory filing cited by The New York Times.”

I know of one public filing that names a private fund’s prime brokers, and I could not find a second. It is Form ADV, Schedule D, Section 7.B.(1), Question 24. I pulled it.

Situational Awareness LP, CRD 333011, SEC file 801-132039, filed 11 June 2026, signed under penalty of perjury by chief compliance officer Kimberly Summe. Question 24(a) asks whether the private fund uses one or more prime brokers. “Additional Prime Broker Information: 4 Record(s) Filed.”

Four names. I count the same four. No addition and no omission. The reported list stops exactly where the filing stops. Morgan Stanley, a top tier US prime broker, appears on neither.

The subpoena list is a filing, not a finding. That is the first thing the documents say. The second is stranger, and I had to read it twice.

A registration record less than half the size of the fund’s own 13F

Question 11 of that same Form ADV asks for the “current gross asset value of the private fund.” Filed on 11 June 2026, the answer is $9,278,344,000. Item 5.F reports identical regulatory assets under management, across two accounts.

Nineteen days after that signature, at the 30 June quarter end, the fund’s own Form 13F reported $20,242,292,228 in US listed long equity across 26 positions. That comes straight off the cover page, and I did none of the sums in it. I make the ratio 2.1817, so 2.18 times, and the gap $10.96 billion. A gap between what an adviser files and what anyone prints is not unique to this fund. I took the same two fields apart on Citadel, where the filed number and the circulated number differ by a factor of eight.

Both figures cover the same book: the ADV reports a single master-feeder fund, the 13F the adviser’s discretionary holdings. Two limits on the comparison, and both cut the same way. A 13F reports long US listed positions at a quarter end and nothing else, no shorts, no derivatives, no non-US holdings, and it lands 45 days late, so $20.24 billion understates the book. Regulatory assets under management is a gross figure, deducting no debt and no leverage, which is precisely why it should be the larger of the two. Both corrections widen the gap. Pull the whole 13F series off EDGAR and the reason becomes obvious.

Each row is long US listed equity at that quarter end, and each is the cover page total, not my sum of the holdings.

$2.12 billion to $20.24 billion in four quarters. I make that 9.53 times. That $9.28 billion on the ADV sits between the December and March rows, which is roughly where a fiscal year end gross asset value would land for a fund whose December long book was $5.52 billion. So the number is stale. Here is the part that matters. It was allowed to be stale. The Form ADV General Instructions, Instruction 4, on what an adviser must refresh when it files an interim amendment:

Item 5 is regulatory assets under management. Item 7 is private fund reporting, which is where Question 11 lives. The 11 June filing is captioned, in its own header, “Other-Than-Annual Amendment - All Sections.”

I re-checked that filing on 26 August, six weeks after the fund reportedly lost around $35 billion, an unwind I traced through Citadel’s own subsequent selling. It is unchanged. Nobody did anything wrong here. That’s my whole point. “Current” in Question 11 means current as of the last time the form was obliged to carry a number, and for a fund that grew ninefold inside a year, the obligation arrives once. A regime that updates annually cannot see a fund that moves quarterly. No rule was broken and the record was wrong anyway.

If you want to run this on your own book. I pulled Question 24 for six advisers this week, this fund among them, to see whether naming four prime brokers is unusual. The census, the two step recipe for pulling that field on any SEC registered adviser, and what the spread does and doesn’t tell you are in the members’ note on Patreon.

Read more

The $14.6bn refinancing everyone priced in August was not Jane Street’s largest new fixed charge of 2026. A fifteen year absolute triple net lease on an Oklahoma data centre, signed the next day and parent-guaranteed, commits it to $289m to $311m a year against roughly $175m of incremental interest on the notes. The notes are counted, rated and public. The lease is none of those.

Below the paid line: the committed charge arithmetic, 149MW at the EIA’s June 2026 Oklahoma tariff of 7.37c/kWh, to a fifteen year undiscounted $4.34bn to $4.66bn. Moody’s 30% trigger worked line by line at 26.5% and 30.6%, and the 8x-rent method that lands it at 29.5% instead. Then the full 2026 obligation stack, the capacity table, the ICE BofA control, and five confounds I cannot rule out.

What Jane Street’s $14.6bn refinancing bought, and why it read like the whole story

On 12 August Jane Street priced $14.6bn of senior secured notes into a private group led by JPMorgan with PIMCO in the book: $5.86bn to 2031, $5.13bn to 2033, $3.64bn to 2036, the long piece at 8.088%. Roughly $11.2bn retired a $5.5bn term loan and $5.7bn of existing notes, as Bloomberg reported. The reading that followed was coherent and mostly right. New money costs about 311bp against roughly 152bp on the paper it replaced, about $175m a year of incremental interest, and it takes the funding private. Bloomberg Opinion called the swap smart self-defense. A firm that had just printed its first losing month in a decade paid nine figures a year to control who reads its numbers. I wrote about that two weeks ago and still think the disclosure motive is right. This piece is about the deal nobody set beside it.

The three obligations of 2026, and the one that got priced

Jane Street has been buying compute all year, four ways, and nobody has put the year in one table.

CoreWeave’s announcement is unambiguous: “Jane Street has committed approximately $6 billion to use CoreWeave’s AI cloud platform,” alongside a $1bn equity investment at $109.00 a share. It states no term, a gap I return to. Bloomberg reported in June that the firm is separately weighing its own 100MW to 200MW data centre, self financed, on a plan to go from tens of thousands of GPUs to hundreds of thousands.

The two off metric lines come to $8.25bn in one year, against $1.90bn of headroom under Moody’s test, worked below.

⛔ Those two are not the same instrument. A guaranty of third party debt is close to debt and is the kind of contingent obligation a rating methodology can capitalise. A cloud purchase commitment is a vendor contract, and no agency’s leverage metric counts those as debt, for any issuer anywhere. The lease is the sharp case. CoreWeave is not evidence of a gap in anyone’s metric. That one costs me the headline number. It’s there for scale: its April date turns the August lease into a pattern.

Zenith Arc, 149MW, and a lease with no bid

Zenith Arc LLC, a subsidiary of Prairieland Rivers LLC, sold $2.25bn of green bonds due 2031 to fund a data centre and substation in central Oklahoma: 149MW, leased for fifteen years to JS Data Center LLC, a Jane Street Group subsidiary, with the parent guaranteeing. That lease is no normal tenancy: under an absolute triple net lease with a guaranty Jane Street owes base rent and, separately, every property expense: maintenance, property taxes, utility bills, insurance premiums, and electricity. All of them, uncapped, fifteen years, whether the machines inside earn anything or not. Moody’s rated the notes Ba2, two notches below investment grade. I read the guaranty as why they cleared at all.

That shape is why the credit prices where it does. A firm at 6.2x balance sheet leverage borrows at investment-grade-adjacent levels because its book is self-liquidating: if the business stops on a Tuesday you sell it and pay everyone back by Friday. Jane Street Capital’s own Form X-17A-5 for 2025 shows it: assets $60.70bn, liabilities $54.42bn, equity $6.28bn, and securities sold short of $45.01bn against long positions of $41.70bn. A book made of things with bids. A fifteen year lease on an Oklahoma substation has no bid, cannot be sold or hedged, and does not shrink when the trading book does. July was the first down month in a decade; the rent would have been due anyway. Jane Street has always had office leases, so non trading obligations aren’t new. The scale is, and so is a public rating letter on one.

Baa3, Ba1, Ba2: Moody’s three notches, and why they prove less than they look

Moody’s affirmed Jane Street Group at Ba1 on 22 July, outlook to positive, and affirmed Baa3, investment grade, on the four wholly owned operating subsidiaries. Two days later Fitch upgraded the group to BBB- from BB+, citing “strong through-the-cycle growth in net operating income.” Three weeks later the same agency looked at a bond whose economics are almost entirely Jane Street’s promise to pay rent, and rated it Ba2.

I wanted that ladder to mean the market prices Jane Street’s duration. It doesn’t, and I’ll say why before someone else does. Clifford Chance’s July briefing on exactly these structures says data centre lease bonds are “notched down for structural reasons (e.g., completion risk or refinancing risk at a bullet maturity)”, with the notch being about structure, not credit. Zenith Arc has both: a construction financing with a 2031 bullet against a fifteen year lease. And S&P didn’t notch it at all, assigning a preliminary BB, where it already rates Jane Street.

The ladder is real. It is not evidence of what I wanted, and I am not going to dress that up. What survives: a Jane Street obligation is now rated in public, on its own, by two agencies, and neither puts it above the firm. The rest runs on the cash.

The rent floor is $202.5m. The power bill is bigger than I expected.

Nobody discloses the rent, so I built a floor from the debt it services. $2.25bn priced at 99.5 to yield close to 9%. That is $202.5m of interest a year, and under an absolute triple net structure rent must cover it. Real rent sits higher: it carries the developer’s equity return, fees and reserves too.

Electricity is where I was wrong. I expected a rounding line; on 149MW of critical IT load it isn’t, so I priced it. The EIA’s Electric Power Monthly, Table 5.6.A puts Oklahoma industrial electricity at 7.37 cents per kWh in June 2026 against 6.57 a year earlier, a tariff rising 12% a year that the tenant absorbs uncapped. Two cases, bracketing the uplift for cooling and losses:

Add them. I make the committed annual floor $289m to $311m: $202.5m of rent plus $87m to $108m of power, before the property taxes, insurance and maintenance the lease also assigns to Jane Street and which public data cannot size. Fifteen years, undiscounted: $4.34bn to $4.66bn, or 69% to 74% of the $6.28bn of equity in the broker-dealer that does the trading. Now put the two August deals side by side.

The obligation nobody wrote about runs between 1.65 and 1.78 times the size of the one everybody did. One caveat, because it isn’t like for like: $175m is the INCREMENTAL cost of the refinancing while $289m to $311m is the lease’s GROSS charge. Incremental is the right basis here because the subject is what Jane Street ADDED in 2026: the refinancing added $175m, the lease at least $289m. My load and PUE assumptions are moveable and a reader should move them, though not far enough to reverse this: set the power bill to zero and the rent floor alone still exceeds $175m. That’s the gap.

Read more

On 19 August the Treasury said it would increase, “by at least double,” the size of its liquidity support buybacks in the ten to twenty and twenty to thirty year sectors, from $2 billion per operation to “at least $4 billion.” A liquidity support buyback is Treasury paying cash to dealers for old, illiquid bonds it already owes on, at prices the dealers name and Treasury can refuse. The change is effective 9 September and runs through the 4 November refunding (Treasury). Within five days two of the most serious people in this market had called it a mistake.

They’re both right about the intent. I think they’re both looking at the wrong side of the trade, and there’s a paper that says the mechanism is a third thing again.

Below the paid line: the full 22 operation table for 2026 printed row by row · the constant cap experiment, five sectors at an identical $4bn ceiling with cover from 1.3x to 11.0x · Nohshad Shah’s 13% recomputed against Treasury’s own auction table, and the IMF paper underneath the argument · the three readings of the 10 September print, each with the offer number that distinguishes it · the two operations where Treasury refused to fill, and what the curve did the next morning.

Druckenmiller argues dysfunction. Shah argues duration.

Stanley Druckenmiller ran the argument from dysfunction. His Wall Street Journal opinion piece of 24 August, behind that paper’s paywall, says there were “no failed auctions, no dealer balance-sheet seizure, no forced unwinds,” and that trading was orderly, “not a malfunction but the machine doing its job.” I’m quoting it as reported (Axios), not from the original. His conclusion follows cleanly. With no dysfunction to fix, the operation was price management wearing a liquidity label.

Nohshad Shah at Citadel Securities ran it from duration. His note is public and worth reading in full (Citadel Securities). One thing worth clearing up, since the obvious cheap shot is wrong: Citadel Securities is not among the twenty five firms on the New York Fed’s primary dealer list, so it cannot offer into these operations at all. Whatever else his note is, it isn’t a man talking his book into a window he stands at.

I wrote up the doubling itself when it was announced, and that arithmetic still holds (Bessent Doubled the Buyback).

Shah counted seven operations left on the schedule, put the incremental capacity at “at least $14bn,” and called that “around 13% of expected quarterly 20y and 30y issuance.” He declined the easy QE framing, because purchases financed by bills leave the debt stock where it was. “The cleanest framing is a Treasury-led Operation Twist, broadly liquidity neutral, but nowhere close to duration neutral.” Financial repression, in his phrase, “at the margin.”

I recomputed both halves before leaning on them. Treasury’s August refunding statement sets twenty and thirty year auction sizes at $16bn and $25bn for August, then $13bn and $22bn for September and October. That totals $111bn, and fourteen over 111 is 12.6%. His 13% holds, and the tentative schedule gives exactly the seven operations he counted.

Both men did the work. My problem is that neither opened the file that answers the question they were arguing about.

The offer column: $520 billion submitted, $42 billion accepted

I expected the interesting number to sit on Treasury’s side of the ledger, because that is the side both critiques argue about. It isn’t there. The number I wanted sits one column to the left.

Every buyback is two sided and Treasury publishes both sides. Dealers submit the par they wish to sell, at prices they choose; Treasury announces what it took. That offer number costs nothing, publishes the same day, and runs back to 2024.

Here is every long end operation of 2026. Twenty two of them, offered against accepted, with cover as a multiple of the cap Treasury set itself.

Read the accepted column first. Twenty one of twenty two print $2.000 billion, to the dollar, which is the cap. The exception is 19 March and I come back to it.

Across the twenty two, dealers offered $520.4 billion and Treasury took $42.2 billion (Treasury buyback operation results). I make that 8.1%. Be careful what you take that number to be. It measures what dealers wanted off their books, at prices they named themselves, and Treasury’s role is to decline 92% of it. That’s the whole finding.

This data is not a secret. Specialists who follow buyback plumbing track it, and one shop publishes a live dashboard whose headline metric is the offer to cover ratio. It appears in neither critique.

Druckenmiller picked the right test and this is where it gets answered. He went looking for a seizure and was correct that there wasn’t one. A full dealer balance sheet and a broken one are different conditions, and only the second shows up in auction tails and realised volatility. The first shows up here.

The one month bucket draws 11.0x, and it killed my duration story

My first draft said dealers offer where duration hurts. Then I read my own table.

Treasury ran liquidity support operations in five other coupon sectors this year at a cap of $4 billion, twice the long end ceiling. Same mechanism, same counterparties, same pricing discipline on Treasury’s side.

Look at the top row. The one month to two year bucket has almost no duration and it draws the highest cover in the table. Duration cannot be the explanation. A reader would have caught that in four seconds. I nearly didn’t.

The variable that fits all six rows is illiquidity and carry cost. At the front, off the run paper near maturity is odd lot inventory nobody wants to warehouse. At the back it’s off the run long bonds with thin two way flow and heavy balance sheet cost. Two different reasons. One behaviour. Dealers offer where holding hurts.

I’d have written that as my own inference, except somebody already proved it. Jing Zhou’s IMF working paper Testing the Liquidity Support Effects of the U.S. Treasury Buyback Program instruments Treasury’s own selection rule and finds buybacks “reduce primary dealers’ net holdings of Treasury bills and coupons,” modelled as “predictable demand for dealers facing inventory constraints and holding costs.” That is the mechanism, identified properly, a year before this argument started. Somebody got there first.

One more tell sits in the same file: Treasury accepts a median of three CUSIPs out of thirty six eligible, picking the cheapest against its own internal value and sending the rest home. Three of thirty six.

Read more

Scott Bessent doubled the ceiling on Treasury’s long end buyback operations on 19 August, and within a day the framing settled: Treasury is suppressing long yields, Kevin Warsh is fighting inflation and shrinking the balance sheet, and the two are colliding.

I think the collision everyone’s describing doesn’t exist in the flows. Something larger does, and it sits at the opposite end of the curve.

Below the paid line:
- The Desk’s full month by month bill purchase table, December 2025 to September 2026, $355 billion planned, and the moment reserve management purchases went to zero
- The H.4.1 line by line, both dates, showing bills at more than 100% of the growth in the Fed’s Treasury book
- The TGA at $933.2 billion and the Desk sizing rule that makes Treasury’s cash balance an input to Fed open market operations
- The TBAC charge, in which Treasury’s own advisory committee prices $540 billion of Fed bill buying and tells Treasury it can raise the bill share without the private market absorbing it
- The 2000-02 buyback programme that moved yields 95bp, and why its budget-surplus funding means it cannot repeat against $739 billion of net new borrowing
- Three confounds I cannot rule out, including the one that would most weaken this piece

The consensus, stated the way its believers would state it

Here’s the strong version, and I want to be fair to it, because it isn’t stupid.

Treasury finances the government at least cost and isn’t supposed to have a view on long rates. When it announced it was increasing “by at least double” its liquidity support buybacks in the 10-to-20 and 20-to-30 year sectors, it was announcing a view. Bessent has said for two years that he cares about the 10-year, and the 30-year had just printed above 5.25%.

Meanwhile Warsh has told reporters that rising long-term borrowing costs served a valuable purpose, tightening financial conditions with inflation at 3.4% in the year to July against a 2% target. So one arm of the state unwinds what the other relies on. Rebecca Patterson at the Council on Foreign Relations called the buybacks “more signal than substance”, noting real QE is unavailable because Warsh “opposes sustained balance sheet expansion.”

That’s a coherent story and every element is checkable. I think two are wrong.

What the schedule actually says

Treasury publishes its tentative buyback schedule every refunding. I pulled the Q3 2026 file and counted the operations inside the doubling’s 9 September to 4 November window. There are seven long-end nominal operations: 10 and 24 September, 1, 8, 15 and 27 October, 4 November. I make that eight weeks. Seven operations at a $4 billion ceiling is $28 billion of maximum capacity, against $14 billion at the old ceiling. So the doubling buys Treasury $14 billion of extra room across eight weeks, in a market where one August refunding sold $42 billion of 10-year notes and $25 billion of 30-year bonds.

Two details matter more to me than the headline. First, the minimum purchase amount column reads $0 on every line, so I read these as ceilings, not commitments: Treasury buys what dealers offer at prices it likes and is obliged to buy nothing. Second, the sunset. The doubling runs through 4 November 2026 and then expires, and a permanent programme of suppressing yields doesn’t come with an eight week expiry printed on the announcement. That sunset matters.

One detail nobody mentioned: the cash management buybacks on 3 and 9 September carry ceilings of $12.5 billion each, in the one-month to two-year sector. Treasury’s biggest scheduled buybacks this quarter are three times the long-end ones everybody is arguing about. Front end again.

The Fed stopped shrinking nine months ago

I went looking for the other half of the collision, expecting a balance sheet shrinking while Treasury bought. It isn’t there. So I pulled both H.4.1 releases and differenced the lines myself. The balance sheet grew.

Here the consensus isn’t just imprecise. It’s inverted.

The FOMC announced on 29 October 2025 that it would “cease the runoff of its securities holdings starting on December 1, 2025.” It directed the Desk to “roll over at auction all principal payments from the Federal Reserve’s holdings of Treasury securities” and to “reinvest all principal payments from the Federal Reserve’s holdings of agency securities into Treasury bills.”

I’d read that second clause slowly. Every dollar that comes back from the Fed’s mortgage book gets recycled into T-bills. The Fed’s agency portfolio throws off principal continuously as mortgages amortise and prepay, so it’s a standing, mechanical bid in the bill market that needs no further decision from anybody.

Then the Committee went further. Its 10 December 2025 directive ordered the Desk to “Increase the System Open Market Account holdings of securities through purchases of Treasury bills,” and to “Reinvest all principal payments from the Federal Reserve’s holdings of agency securities into Treasury bills.” The Desk implemented it as purchases in the secondary market. These are reserve management purchases, and the Desk’s own statement says the first schedule totalled approximately $40 billion in Treasury bills, starting 12 December 2025.

For context: in the period ending 11 December 2025 the Desk ran no Treasury operations at all, and through 2024 and most of 2025 it alternated between nothing and $150 million test operations. Zero to $54 billion a month. That is a regime break.

I pulled the Desk’s full monthly schedule and added it up. Planned bill purchases from 12 December 2025 through 14 September 2026 come to roughly $355 billion, split about $140 billion of agency reinvestment and $215 billion of reserve management purchases.

The balance sheet confirms it, line by line

Schedules are only plans. The H.4.1 is the receipt, and it settles the question.

Source: Federal Reserve H.4.1, both dates. I’d read that bottom row twice.

That balance sheet grew $127 billion over the year in which Warsh was supposed to be shrinking it. Its Treasury holdings grew $340 billion. And the bill line grew $342 billion. That is more than 100% of the growth in the whole Treasury book, because the coupon book barely moved and the inflation-indexed book fell. Bills did all of it.

Bills went from 4.65% of the Fed’s Treasury book to 11.84% while the mortgage book shrank $186 billion. The Fed has been converting mortgage duration into Treasury bills, at scale, on autopilot, under a directive issued before Warsh took the chair. Nobody voted on that this year.

So I’d say the sentence “Warsh is shrinking the balance sheet while Bessent buys bonds” fails in both halves. Warsh isn’t shrinking anything, and Bessent’s long-end buying is capped at $28 billion through early November.

The mechanism: where the two balance sheets actually touch

Here’s the part I think is mispriced, and it has nothing to do with the 30-year.

Treasury’s cash sits in the Treasury General Account at the Fed. The TGA is a Fed liability competing with bank reserves, so when it falls reserves rise dollar for dollar and when it rises they drain. It’s a monetary lever Treasury operates and the Fed doesn’t. That account held $933.2 billion on 24 August, per the Daily Treasury Statement, and Treasury assumes near $950 billion at end-September. Reporting suggests Bessent has looked at drawing it down.

Now read the Desk’s sizing rule again. Reserve management purchases are “sized to accommodate projected trend growth in the demand for Federal Reserve liabilities as well as seasonal fluctuations.” The TGA is one of those liabilities, so Treasury’s cash decisions are a direct input into how many bills the Fed buys. I’m not inferring intent. That’s the Desk’s own published methodology. The rule is published.

This year, reserve management purchases ran at $40 billion a month through the first quarter, stepped to $25 billion in April, then $10 billion, and for 14 August to 14 September the Desk plans none. The mechanical reinvestment leg has climbed from $13.4 billion to $17.0 billion and is now the whole programme. Discretion off, autopilot on.

So the discretionary leg is off. The automatic leg is at its highest level since the directive began, and the Fed is still buying $17 billion of bills a month.

The honest comparison is less flattering than the one I first wrote. Treasury’s seven operations at a $4 billion ceiling run across 56 days, a maximum of about $15 billion a month. Per month the two are roughly the same size. The asymmetry is duration, not rate: the buyback is an eight-week ceiling with a $0 floor expiring 4 November, and the Fed’s bid is standing, nine months old, and needs no decision to continue.

Treasury issues bills at the margin, and its refunding statement says so: it meets variation in borrowing needs “through changes in regular bill auction sizes and/or CMBs.” The Fed is a structural buyer of bills, sized partly by a Treasury-controlled liability. Neither institution decided to build that loop.

Read more

The claim: The SpaceX job posting everyone read as fuel procurement asks for profit and loss generation, pricing inefficiencies and CFTC compliance, which is a merchant desk specification, and it makes SpaceX the fourth buyer of this size to become an energy trader after Amazon, Google, Microsoft and Meta. The hardware that made it necessary is already paid for: a 5% stake carried at $7.2m paid out $60.4m in under seventeen months, pricing the mobile turbine fleet at $1.21bn against the $1bn that got reported.

The numbers: $50,392,931 of cash plus a $10,013,872 holdback to a 5% holder, and 850 MW across 30 mobile turbines (Duos 10-Q, 19 Aug 2026). $1,421 per kW implied, against $2,000 to $2,500 for new build you cannot take delivery of before 2030.

The catalyst: the holdback runs 12 months from 26 May 2026 and settles the price question in a public filing around May 2027.

Wrong if: the holdback is consumed by indemnity claims, or the distribution to Duos was not pro rata, in which case the reported $1bn is closer to right than my $1.21bn.

The consensus: SpaceX is integrating backwards into its own fuel chain

SpaceX needs power on a schedule no utility can meet and methane on a schedule no supplier is set up for, so it is integrating backwards into both.

That version is coherent and I want to give it full weight. The company committed $16.8bn in early August to a first phase of Terafab in Grimes County with Tesla and Intel. At a county public meeting on 6 August, Riley Trettel, who runs energy and data centre development for SpaceX, put it in five words: “We’re bringing our own power.” Gas fired plants, he said, and very large battery arrays. Starship’s Raptors burn liquid methane. ERCOT interconnection queues are long, turbine order books are longer, and a fab that flickers scraps wafers. A gas trader is what you hire when you’ve decided to own the fuel chain. I’d have believed that too, right up until I read the responsibilities.

The posting asks for P&L, and procurement desks do not have one

I went looking for a procurement story. The posting doesn’t contain one. Almost every outlet described the SpaceX listing without quoting it, so I read it myself. The successful candidate will “identify pricing inefficiencies and manage profit/loss across various market segments” and “manage trading positions based on fundamental and technical market signals.” Preferred experience includes “storage optimization” and a “track record of P&L generation.” The role must “ensure compliance with CFTC/FERC regulations.” Familiarity with Endur is listed as a plus. I read each of those as a tell. A company that only wants to buy gas doesn’t manage positions on technical signals. It doesn’t need a P&L, because its objective is a low and stable delivered cost and a P&L invites trading against that. It doesn’t need CFTC compliance, which attaches to derivatives a pure physical buyer never touches. And it doesn’t need Endur, the trade capture system merchant energy books run on. That’s a merchant desk. I don’t think it’s a close call. Storage optimization is the one I’d point a PM at first: buy gas when the curve is in contango, inject it, sell it forward at the spread less carry. It’s a spread business with a term structure view. Nobody does that to keep a chip fab lit. Alcoa hedges power and airlines hedge jet fuel, so the rival reading is ordinary hedging in modern dress. Two things kill it: hedgers use bank counterparties and confirmations rather than Endur, and CFTC compliance attaches to derivatives a pure physical buyer never touches.

One detail I can’t place: the role requires Air Force background checks and up to 50% travel, which points at a physical delivery footprint on a federal launch range, though it may just be the badge rule at that address.

Meta filed to become a power marketer, and SpaceX is the fourth one in

The obvious objection, and it is the right one: why is a job ad news when every large buyer is doing this?

It’s a fair hit. Amazon, Google and Microsoft are already active power traders on their own regulatory filings. I looked at how a dedicated commodity desk actually generates that edge in InCommodities and Danske Commodities, which is the shape SpaceX is hiring toward. In September 2025 Meta went further, filing a market-based rate tariff at FERC through a wholly owned subsidiary, Atem Energy LLC, formed in the application’s words to act as a power marketer selling “energy, capacity, and certain ancillary services” at wholesale across six ISOs.

So SpaceX is not first. I read it as the fourth buyer of this size to stop being a customer of the energy market and start being a participant in it, and the pattern is what matters here, more than the job ad itself. Four in eighteen months.

Two things make the SpaceX version different, and both are about Texas. Meta’s filing names six ISOs and ERCOT is not one of them, because ERCOT is not FERC-jurisdictional. And Meta filed to sell POWER, while SpaceX is hiring to trade the fuel, one layer down, in the only large US grid where fuel and power fail together.

I should be straight about the evidence. A job posting proves what an employer asks for, never what it does. My whole reading of SpaceX’s intent rests on one document, and a recruiter adapting a merchant trading template would produce the same words. What persuades me anyway is that Endur and CFTC compliance don’t migrate into a procurement template by accident, and that the hardware was bought fourteen months earlier.

Duos Technologies’ 5% stake prices the fleet at $1.21bn

Here’s where the private company becomes readable, and I get there through somebody else’s balance sheet. The sequence matters, so take it one date at a time.

On 31 December 2024 Fortress Investment Group closed on 850 MW of mobile turbines. On 14 May 2026 the FTC issued an antitrust early termination notice, transaction 20261350, which reporting ties to Elon Musk. On 26 May 2026 substantially all of those assets changed hands. On 6 August 2026 SpaceX committed to Terafab and said it would generate its own. On 21 August 2026 the gas trading role was posted. Fourteen months from turbines to trader.

A Jacksonville microcap called Duos Technologies Group is how that first date is readable. (I have used SpaceX’s own filings this way before, on its first AI profit and the depreciation add-back underneath it.) It helped Fortress buy, in the filing’s words, “approximately 850 megawatts of electrical generation capacity (consisting of 30 mobile gas turbine generators)” plus balance-of-plant from Atlas Corporation and APR Energy Holdings, through a vehicle later renamed New APR Energy, LLC. Duos took an asset management agreement and a 5% non-voting interest in the parent, carried at $7,233,000.

One line explains how a fleet like this becomes available at all. Duos discloses that “Chuck Ferry, our then Chief Executive Officer, was formerly the CEO of APR from 2018 to 2020.” The microcap that helped Fortress buy APR’s turbines was run by the man who had run APR. Assets like this don’t sit on a screen.

Duos filed an 8-K two days after closing and the press backed a valuation out of it. The 10-Q gives the schedule exactly:

So a 5% holder realised $60,406,803. Divide by 0.05 and I make that $1,208,136,060.

The reported $1bn is $50.4m of cash divided by 5%, and it quietly drops the escrow. That escrow is no maybe in Duos’s own accounting: the company booked it as a receivable, ran it through the gain, and told the SEC anything left after twelve months gets distributed. The gap is $200,277,440, or 16.6% of the larger number. The outlet that ran the figure even showed its working, saying the value was “backed out from disclosures showing a minority stakeholder received roughly $50.4 million for a 5% stake.” The escrow was in the open and left out anyway. Nobody recomputed.

Careful about what that licenses. Grossing up a distribution to a non-voting holdco member measures equity proceeds and assumes a flat waterfall. Duos supports the assumption, calling the $60m the realisation of its 5% interest, but debt, fees or a preference would move the true consideration. So I’d call it $1.0bn to $1.21bn, with the circulating number at the bottom.

The return is cleaner and I have more confidence in it. $7,233,000 became $60,406,803 in under seventeen months. I make that 8.35 times money, or 6.97 on cash alone. Duos ended the quarter with $112.3m of cash against $15.5m at year end and reported $1.58 of basic EPS, for a company that lost money from operations.

Fortress bought turbines at the end of 2024. Musk bought a clock.

$1.42m per megawatt buys a delivery date, and the trade does not repeat

Divide $1,208,136,060 by 850 MW and I get $1.42m per megawatt, about $1,421 per kW. Across 30 units that’s roughly $40m per turbine.

Price the alternative. Recent US combined cycle projects report installed costs moving toward $2,000 to $2,500 per kW, and money is not the constraint. GE Vernova’s backlog ran 80 GW in December 2025 and closed the second quarter of 2026 at 116 GW, with reservations now taken for 2031 delivery.

So I make the trade 57% to 71% of new build cost, for capacity that exists today instead of 2031. Time was the product. Not steel. The buyer did not purchase generation, which anyone can order with five years to wait. He purchased the five years.

Two things anchor it. APR Energy sold to Atlas Corp for $750m in February 2020. That was the whole company. Six years later an 850 MW slice of its assets implies $1.21bn. And the immediacy is not my inference: Fortress’s own announcement of the 2024 purchase called the portfolio “immediately available for deployment.”

One thing I can’t give you is the fab’s own requirement. SpaceX has not disclosed Terafab’s load, so every figure here rests on the 850 MW that was actually bought.

Then ask whether it repeats. I think not. ERCOT’s large load queue went from 63 GW in December 2024 to near 230 GW by March 2026, three quarters of it data centres, against a fleet serving about 105 GW. Published counts run 226 GW to 410 GW depending on duplicate treatment. Against that, Aggreko, the largest operator in mobile power, ran about 9.6 GW worldwide in 2021 and the whole global power rental industry turned over roughly $11.7bn in 2025. APR’s 850 MW is under a tenth of one Aggreko. That is the entire pool.

So the world stock of fast deployable thermal capacity sits in the low tens of gigawatts against one grid’s queue of over two hundred. Nobody publishes that stock figure and I should say so: I built it from a dated fleet number and an industry revenue line, and it’s the least evidenced thing here. It would have to be wrong by an order of magnitude to change the conclusion.

This was an inventory arbitrage on a fixed stock, available once, to a buyer who could write a cheque with no board process and knew which yard the turbines sat in. A fund can’t replicate it. I have written the desk-level version of this section as a separate note: the capacity bound sized, the regime conditions that hold it, and the three risks with probabilities and early warning signs attached. The decay is visible too: GE Vernova is taking annualised production from roughly 20 GW toward 24 GW by 2028, and when OEM throughput catches the queue the immediacy premium compresses and the $1.42m print goes with it. Two to four years, and it closes from the supply side. I’d not bet on demand cooling. Size the other half in the same units: if used mobile capacity reprices toward refurbishment value instead of 57% of new build, an 850 MW fleet bought at $1.21bn is worth a few hundred million less, and the buyer still holds turbines burning 50% more gas per MWh than the plant he could have ordered. That’s the cost of being early.

850 megawatts of aeroderivative turbines burn 59 Bcf. Starship burns 5.4.

Every headline framed this as rocket fuel. The arithmetic doesn’t support that, and I’d sooner show the working than assert it. The APR fleet is 20 GE TM2500s and 10 Pratt & Whitney FT8 mobile packs. TM2500s run 8,830 to 9,705 Btu/kWh; the FT8 starts near 9,018 at ISO and degrades toward 10,400 in heat. Blended, in Texas, I’ll use 9,700 Btu/kWh and assume an 85% capacity factor. Both are mine, so flex them: at 8,830 the fleet burns 53.9 Bcf, at 95% it burns 66.2. The conclusion survives every version.

850,000 kW × 8,760 h × 0.85 = 6,329,100,000 kWh. At 9,700 Btu/kWh and 1,037 Btu per cubic foot I make that 59.2 Bcf a year, about 162 MMcf/d.

The same electricity from a modern combined cycle plant at 6,400 Btu/kWh is 39.1 Bcf a year, so the penalty for choosing mobile simple cycle machines is 20.1 Bcf a year. Against that, the rockets. Super Heavy carries 3,400 t of propellant and a Block 2 ship 1,500 t, so a stack is 4,900 t at a 3.6:1 oxidiser-to-fuel ratio, which I make 1,065 t of methane per launch. At 55.5 MJ/kg a tonne is 52.6 MMBtu, so a launch is 54.0 MMcf. At a hundred launches a year, far above any cadence yet flown, Starship needs 5.40 Bcf. That counts launches only, so it excludes static fires, proof tests and scrubs, and the true rocket number is higher.

Eleven to one.

The turbine fleet burns roughly eleven times the gas the entire Starship programme would at a hundred flights a year, and the efficiency penalty alone is 3.7 times the rockets. A gas desk sized for launch operations is small. 162 MMcf/d is a real one: throughput comparable to a mid-size marketer, and the reason the posting asks for a team.

One caveat I’m not burying: rocket grade methane is liquefied and purified far beyond pipeline spec, so this is an energy equivalence and not a substitutability claim.

Waha went to minus $6.34 and back to plus $2, and a pipeline did all of it

Here’s the objection I’d raise. If the point is fuel security, why build a desk where gas has been free? Through spring 2026 the Waha hub in West Texas ran below zero for a record ninety consecutive days, hitting minus $6.34/MMBtu on 12 March. Between January and mid-June it cleared zero on eleven days, most during a January freeze off. Producers were paying to have gas taken away. You don’t staff a trading floor for that.

Then somebody built a pipe.

On 16 June, Kinder Morgan’s compression expansion on Gulf Coast Express added 570 MMcf/d of takeaway, and Waha turned positive the same day at plus 42 cents. It averaged around $1.64 in July and has run above $2.00 in August, with Hugh Brinson and Blackcomb still to commission fully, while Permian production set a record 24.9 Bcf/d. The basis just went away.

Five hundred and seventy million cubic feet a day of steel erased a seven dollar basis. No demand shift. No production decline. Nothing to do with weather. One compressor station.

That is the argument, run as a natural experiment. The molecule was never the scarce thing. Transport was. The gap between a wellhead giving gas away and a turbine 400 miles east that needs it at 3 a.m. in January is pipe, storage and firmness, and each trades as a basis position instead of a purchase order. Which is why the posting asks for pipelines, storage optimisation and counterparties in the same breath as P&L.

The tail is not theoretical. The FERC and NERC report on the February 2021 freeze found 1,045 units carrying 192,818 MW suffering 4,124 outages, derates or failures to start. Freezing caused 44.2%, fuel issues 31.4%, and gas supply was 87% of that fuel bucket, which is 27.3% of every outage in the event. Texas, Oklahoma and Louisiana production fell 21 Bcf/d at the 17 February low. The report lists the causes of wellhead outages as freezing, “loss of power,” and precautionary shut-ins. Gas needs electricity to move and electricity needs gas to be made. Owning both ends of that loop doesn’t break it. It puts you on both sides at once.

And the detail that should worry any owner of thermal generation in ERCOT: 81% of the freeze-related outages happened at temperatures above the unit’s own stated ambient design temperature, across about 63,000 MW. The equipment didn’t fail because the weather beat its specification. It failed inside its specification. Design margin held. The units went anyway.

Texas already decided that owning your own generation does not buy you out

I think this is the most underpriced constraint in the AI power conversation, and it’s statutory.

Texas Senate Bill 6 applies to any single site load at or above 75 MW. Loads interconnected after 31 December 2025 must install equipment letting ERCOT directly curtail them during firm load shed events and emergencies, customers must disclose on-site backup generation equal to or greater than half the proposed load, and ERCOT can require them to deploy that generation, or curtail, when the grid is in trouble.

I read that twice. Building your own plant next to your own load doesn’t exempt you from curtailment in ERCOT. It hands the grid operator a call on your generator.

Self-supply is not sovereignty. It’s a different counterparty.

There’s an exception for critical load industrial customers and critical natural gas facilities, and a chip fab may qualify. No public filing tells me whether Terafab will claim it. The independence people assume only arrives if the site is islanded and never interconnects, which the desk would then supply with no grid backstop at all.

Either way I think the desk gets more necessary. Interconnected, it manages a generator ERCOT can call. Islanded, it’s the only thing between a $16.8bn fab and a cold start. Both roads need a trader.

Five things I cannot rule out, starting with an unnamed buyer

SpaceX files nothing. It’s private, so every statement about its intent here comes from a job posting and press reporting. I’ve leaned on Duos’s filings because they’re the only audited surface anywhere near this.

What Fortress paid in 2024 is not public, and it’s the number that would settle this. If Fortress bought cheap and resold at a markup, $1.42m per MW is a dealer’s spread and not a scarcity price. A reader who has it can price the whole trade against me.

Grossing up may overstate. Holdco debt or an uneven waterfall would make $1.21bn too high. My defence is that Duos itself calls the $60m the realisation of its 5% interest.

The buyer is not named in any filing I can audit. Duos says only “a third party.” Every identification of Musk traces to reporting on an FTC notice I couldn’t read directly. The money is filed and auditable. The name isn’t. I’ve written it as reporting.

The 850 MW may not be Terafab’s. Reporting ties APR to xAI, and this is the weakest link in the piece: I’m joining a turbine purchase to a chip fab across two Musk entities on circumstantial timing, and no filing makes that join for me. If the fleet never reaches Grimes County the burn arithmetic still holds for whatever campus it powers, but the neat line gets a lot longer.

What would change this view: the May 2027 holdback, and two others

Three dated, public, resolvable tests, and I’d hold the argument to them.

May 2027. The 12-month holdback on the $10,013,872 ends and Duos has to account for it. Full distribution supports $1.21bn. Material indemnity claims move it back toward the reported $1bn, and I’d take the correction.

The next two Duos filings. The company put $68.8m into equipment deposits and booked $18.8m of GPU-as-a-service contract liabilities, rotating out of power hardware into compute. If those convert to revenue, the microcap that sold the turbines becomes a read on the same demand curve from the other side.

ERCOT registration. If SpaceX registers Terafab’s generation as an ERCOT resource instead of islanding it, the desk is a market participant and my “he bought a calendar” framing understates the ambition. If it stays behind the meter, the gas book is a cost and risk operation with a P&L bolted on, which is what I currently think it is.

What I would do with it: long firm transport, and watch the $1.42m print

For anyone underwriting the AI power trade, the output is the price. $1.42m per MW is now an observable clearing level for immediately available thermal capacity, against $2.0m to $2.5m for new build delivering in 2030 or 2031. That spread is the market pricing time, and it’s wide enough that I’d expect more distressed and end-of-life fleet M&A before relief from the OEM order books.

For gas, the position is long firm transport and storage into Texas load centres. Waha went from minus seven to plus two because 570 MMcf/d of pipe opened, which tells you which half of that market was ever scarce.

And for anyone holding IPPs or gas marketers on the AI demand thesis, this cohort points the other way. Amazon, Google and Microsoft trade power, Meta filed to market it, SpaceX is hiring to trade the fuel. They’re buying the generation, the pipe and now the desk. That’s demand for molecules and disintermediation of margin, and those don’t net off.

The question I would put to a room of PMs: at $1.42m per MW for used mobile turbines against $2.0m to $2.5m for new build you cannot take delivery of until 2031, which of those two numbers is the mispriced one?

📊 The Decision-Grade Version

This piece is complete on its own: the thesis, the evidence, and what would kill the view are all above, and nothing was held back to sell you a next step.

The Patreon note is a different piece of work, not a deeper cut of this article. It takes the part a desk would actually act on, the capacity bound and the regime conditions around that $1.42m per megawatt print, and writes it the way a desk would use it: what is priced, where the edge stops repeating, the regime that breaks it, and three risks with probabilities and early warning signs attached. Written for people who put capital behind a view.

→ Read the turbine-pricing note

→ Or join the Patreon community for every note

I publish the films at @TheMathematicalTrader and post the working notes on LinkedIn.

Cover photograph: Mobilus In Mobili, CC BY-SA 2.0, via Wikimedia Commons.