NextFin News - Jane Street, the Wall Street market maker that turned artificial intelligence into the most profitable trade of the decade, lost about $15 billion in July — its first monthly loss in roughly ten years — after a concentrated bet on an AI hedge fund and wrong-way positions in Asian equities reversed in a matter of weeks. The loss came even as the firm booked more than $40 billion in net trading revenue for the year through mid-August, a sum that already exceeds its record-setting 2025 total.
The reversal raises a question that reaches far beyond one firm's monthly profit-and-loss statement: when the entire market is positioned for the same structural story, does the risk sit in the trade itself — or in the fact that everyone agreed it could not fail?
The Numbers Behind the Loss
July's roughly $15 billion drawdown broke a decade-long streak of monthly gains for a firm that has become synonymous with quantitative discipline. Turner Batty, a Jane Street partner, acknowledged the result in an internal note to staff:
"July was a bad month," Batty said in the note. "Despite the large year-to-date increase in trading capital, the recency of these losses has caused us to locally be more selective about risk. We've closed a significant portion of our risk in the specific areas we lost on in July, and have also reduced risk-taking in other strategies."
The damage came from two directions. The first was Jane Street's investment in Situational Awareness, the AI-focused hedge fund run by Leopold Aschenbrenner, the former OpenAI researcher whose 2024 manifesto on artificial superintelligence made him a Silicon Valley celebrity. The second was a set of long positions in Asian equities — including non-AI names that had outperformed earlier in the year — that moved against the firm as regional markets cracked.
"We largely lost on the same portfolio of trades that had strong outperformance in the second quarter," the firm said in its internal note. "AI-exposed stocks were down a lot during July, several of the largest memory and semiconductor stocks were down around 50%."
How the AI Trade Unraveled
Situational Awareness was, until July, the clearest proof that the AI thesis still had money to be made. Launched in July 2024 with about $225 million from Stripe co-founders Patrick and John Collison, former GitHub chief executive Nat Friedman and investor Daniel Gross, the fund grew to roughly $45 billion in assets by early July 2026. It returned 439 percent net in the first half of the year, using leverage that reached as high as four times capital.
The book was concentrated long in AI infrastructure — SK Hynix, CoreWeave, Nebius, Micron and Bloom Energy — and short software names including Adobe. Then the trade reversed. The Philadelphia Semiconductor Index fell 25 percent from its June high, a 20.6 percent decline for the calendar month, after posting its best quarter since the index began in 1994. The Nasdaq 100 fell 6.6 percent in July. Many of the fund's largest holdings dropped 35 percent or more while its short positions moved against it, squeezing both sides of the book at once.
Margin calls arrived from Goldman Sachs, JPMorgan and Bank of America. On July 30, before the open, Situational Awareness sold its entire public equity book — a portfolio that had stood at roughly $16 billion — to Citadel in a single block. The fund lost around 67 percent in the month and finished with roughly $10 billion in assets, down from $45 billion at its peak.
For Jane Street, the timing was especially costly. Its stake in the fund had swelled precisely because the fund had performed so well through the second quarter; the drawdown left the position flat for the year, though still profitable over the life of the bet. The buyer on the other side of the unwind was one of Jane Street's closest competitors.
The Hedge That Failed by Design
The most revealing detail in Jane Street's internal account is not the size of the loss but why the firm's hedges did not work.
"We generally worry most about sharp drawdowns, and buy puts that would help in those scenarios," the note said. "The losses in AI stocks were relatively spread out throughout the month, so those short-term hedges provided little help."
This is the mechanics of the failure in one sentence. Convexity — the kind of protection puts provide — pays off when prices gap down violently in a few sessions. It does not pay off when an entire theme is repriced slowly over four weeks. Jane Street's risk book was built for a crash. What it got was a grind.
The distinction matters because it separates an execution error from a design flaw. A firm can survive a crash if it is hedged for one. A firm that assumes its positions can only move in the direction of the consensus narrative has not hedged the narrative itself. The AI trade was not just a position; it was the market's dominant story, and Jane Street's protection assumed the story would end with a bang rather than a slow realization that valuations had outrun earnings.
It Was Not Just One Fund
The damage extended well beyond Jane Street and Situational Awareness. A note from JPMorgan's prime brokerage desk, drawing on Pivot Path data, put July's decline for technology-focused hedge funds at 10.2 percent — the worst month on record for the category, and that figure excludes Situational Awareness entirely. Hedge funds across all strategies gave up almost 3 percent of their 2026 gains in July but remained up around 8 percent for the year. Tiger Global Management's long/short equity fund fell 4.8 percent in July, while Viking Global Investors, which held little AI exposure, lost just 0.2 percent.
The geography of the selloff explains why Jane Street's Asian equity bets compounded the problem. South Korea's KOSPI, the epicenter of the AI hardware rally, triggered market-wide circuit breakers on consecutive days in late July as the index fell below 6,000, after a single-session plunge of more than 10 percent on July 28 sent Samsung Electronics and SK Hynix down double digits. Memory-chip makers, the backbone of the AI hardware supply chain, were at the center of both the rally and the collapse.
In the United States, the semiconductor rout erased a second quarter in which chip stocks had surged more than 80 percent. The pattern was consistent across every measure: the trades that worked in the first half became the losses of the second.
Cyclical Reversal or Structural Reckoning?
The central question for investors is whether July was a cyclical drawdown that will revert, or a structural signal that the AI infrastructure build-out has entered a lower-return phase. The evidence points to a cyclical reversal layered on top of a structural change in how the market prices AI — and confusing the two is the most dangerous error here.
The cyclical case is strong on its own terms. Semiconductor demand has not disappeared: artificial-intelligence workloads still require high-bandwidth memory and advanced packaging, and the supply constraints that powered the rally remain real. Historically, semiconductor cycles mean-revert; sharp corrections after parabolic quarters are typically followed by recoveries once inventory and positioning normalize. July's selloff was driven by leverage and margin calls, not by a collapse in end demand — forced selling of exactly the kind that creates overshoots. When forced sellers are gone, prices often find a floor before fundamentals do.
But the structural leg is equally real, and it is the one the market has not fully priced. For two years, AI infrastructure traded as if the only risk was supply shortage. July introduced a second risk factor: the market now demands evidence that AI capital expenditure is producing returns, not just announcements that it is happening. That is a regime change in the discount rate applied to AI earnings, and it does not self-correct when leverage clears. A stock that was valued on a supply-constraint narrative cannot simply revert to its old multiple when the narrative itself has shifted from "scarcity" to "show me the cash flow."
The practical implication is that a cyclical bounce is likely — positioning is washed out, and forced sellers are gone — but a return to the 2024-2025 valuation regime is not. The mean to which these stocks revert has moved lower.
The Second-Order Risk: When the Winner Becomes the Crowded Trade
The first-order lesson of July is that leverage amplifies losses. The second-order lesson is more uncomfortable: the firms best known for risk management were exposed because the AI trade was not treated as a risk at all — it was treated as the base case. Jane Street's $15 billion loss did not come from a rogue desk or a model failure. It came from a position that grew large precisely because it kept winning, in a market where the winning trade and the consensus trade were the same thing.
This transmission channel runs through the entire hedge-fund industry. When a theme dominates returns for multiple quarters, prime brokers loosen margin terms, volatility stays suppressed, and the hedges that should protect against the theme's reversal become the first cost cut. The result is a system that is hedged for the wrong disaster: violent, idiosyncratic crashes rather than the slow repricing of a shared belief. July proved that a four-week grind can do more damage than a two-day crash when everyone's protection expires before the pain arrives.
There is also a competitive dimension worth noting. Citadel's acquisition of Situational Awareness's book means one of Jane Street's closest rivals now holds the same AI assets at a lower entry price. If the cyclical rebound materializes, the recovery value accrues to the buyer, not the seller. The market did not just take money from Jane Street; it transferred the position to a competitor.
The Counter-Thesis: It's a Blip, Not a Regime Change
The strongest argument against the structural reading is simple: one month does not rewrite a two-year trend, and Jane Street's underlying business remains exceptionally strong. The firm's market-making and short-horizon trading operations — the core engine that handles thousands of trades in milliseconds — were not the source of the loss.
"Market volumes have been strong, and we've continued to make improvements to our short time horizon strategies, that trading seems more profitable than ever," Batty said. "Our positions currently seem appropriate for our present risk tolerance."
This view holds that July was a liquidity event, not a valuation event. The AI infrastructure thesis — that data-center build-out requires exponentially more memory, power and networking — is intact. Under this reading, the firms that held through the volatility, or bought the forced sales at a discount, will be vindicated as earnings catch up to the capital being deployed. Jane Street's $40 billion-plus year-to-date revenue demonstrates that the house is far from burning down.
The counter-thesis is credible on the liquidity point but too quick on the valuation one. A liquidity event becomes a valuation event when it forces the marginal holder to sell — and margin calls are the definition of forced selling. The very existence of a forced block sale is evidence that the market required a lower price to absorb the risk. The structural question is not whether AI infrastructure will grow; it will. The question is whether it grows into its valuation, and July's answer was: not at the old price.
The signal that would prove the structural view wrong is specific and observable: if the Philadelphia Semiconductor Index reclaims its June 2026 high within six months on the back of earnings growth rather than multiple expansion — that is, if revenue and margins, not just sentiment, drive the recovery — then the repricing was cyclical noise and the old regime remains intact. If instead the index stabilizes well below the peak even as earnings improve, the multiple has reset and the structural read stands.
What Comes Next
In the short term, the cleanup is already underway. Jane Street has closed a significant portion of its risk in the loss areas and reduced risk-taking elsewhere. The firm was simultaneously preparing a $14.6 billion bond offering across three tranches to overhaul its debt load and revamp an $11 billion capital stack, with investors including PIMCO, Capital Group and Fidelity participating — a sign that funding markets remain open to the firm despite the loss.
For the medium term, three data points will determine whether this was a one-month anomaly or the start of a lower-return phase for AI infrastructure. First, semiconductor earnings over the next two quarters: revenue growth without multiple expansion confirms the structural reset. Second, hedge-fund positioning data: if technology-focused funds rebuild concentrated AI exposure quickly, the crowded-trade risk rebuilds with it. Third, memory-chip prices and the KOSPI: South Korea's market was the epicenter of the AI hardware rally, and its recovery — or lack of one — will signal whether the supply chain has found a floor.
Scenarios split cleanly. In the base case, AI infrastructure names stabilize at lower valuations, earnings grow into them gradually, and the firms that bought forced sales at a discount outperform the firms that sold. In the upside case, a fresh wave of AI application demand reignites the hardware cycle and the June peaks are retested on earnings rather than enthusiasm. In the downside case, capital-expenditure announcements slow, the earnings gap widens, and the slow repricing that hurt Jane Street in July becomes a multi-quarter compression.
The beneficiaries are the buyers of forced liquidations and the funds that stayed underweight the crowded trade; the exposed are the managers whose best quarter became their largest position. Jane Street sits in both camps — a seller at a loss on one side, a still-dominant market maker on the other.
July's lesson is not that AI was the wrong trade. It is that the most dangerous position is the one everyone agrees cannot lose — because when it does, the exit is owned by whoever has the cash to buy it. Jane Street learned that a hedge against a crash does not protect you from a consensus that slowly changes its mind.
Explore more exclusive insights at nextfin.ai.
