Citadel Just Became the AI Correction's Buyer of Last Resort

On July 30, 2026, Reuters, the Wall Street Journal and the Financial Times reported that Citadel had purchased the bulk of Situational Awareness LP's roughly $16 billion public stock portfolio. The fund, run by 24-year-old former OpenAI researcher Leopold Aschenbrenner, had been down about 67% in July and was forced to unwind most of its broker-financed positions. The deal was brokered through prime brokers including Goldman Sachs, JPMorgan Chase, Bank of America and Citigroup. Per the FT, Situational Awareness retained its private book — anchored by an estimated $5 billion Anthropic stake — and intends to keep managing money rather than wind down.

Citadel's move is structurally novel for this cycle. Through 2025 and the first half of 2026, the AI trade absorbed billions in retail, ETF and fundamental long-only flows without any high-profile forced buyer of last resort. The Aschenbrenner liquidation is the first episode where a marquee AI-thematic fund couldn't meet margin and a major prime broker-faciliated peer stepped in. Per HedgeWeek, Aschenbrenner's investor letter said the firm will stop using bank borrowing to finance positions and will rebuild — a tacit admission that the architecture, not the thesis, killed the trade.

Situational Awareness had roughly 42 positions in its Q1 2026 13F, weighted into AI infrastructure: Broadcom, Nvidia, Advanced Micro Devices, Oracle, Micron Technology, Taiwan Semiconductor, CoreWeave, Bloom Energy and Intel. The fund's 439% H1 return came almost entirely from this concentrated, leveraged long basket. The reversal was equally concentrated: when AI chip names sold off on China-process-progress headlines and a Broadcom Q2 print that, despite a beat, failed to reignite the bid, the same basket that printed 439% in six months gave it back in three weeks.

Two structural features explain why the unwind was so violent. First, leverage: the fund used bank borrowing to magnify the long book, so a modest spot drawdown became a forced-seller drawdown. Second, crowding: per Forbes, the situation resembled a \"bank run\" — once the largest AI-thematic name started selling, every copycat positioning triggered the same liquidation reflex.

Citadel is not a strategic acquirer. The $67B firm (per Wikipedia, January 2026) is the world's second-largest multi-strategy hedge fund after Millennium ($87B). It runs pods across equities, credit, quant and commodities, and trades around 5–7% of US equity volume on a normal day through Citadel Securities. When Citadel buys a $16B public-stock block from a distressed seller, it isn't making an AI thesis call — it's running an inventory and pricing arbitrage against the bid-ask spread it controls. That makes it the ideal buyer of last resort: deep enough to absorb the position without forced selling, fast enough to price it against its own flow.

What changes for AI-stock pricing power: the marginal long has been replaced with a market-maker's inventory. Citadel will run this book on a shorter horizon than a thematic fund would, which means price-insensitive demand has evaporated even as price-insensitive supply has been absorbed. The practical effect — chip names have a higher effective floor (because someone with capital and a market-making edge holds the float) but a lower ceiling on any rebound (because Citadel will not stick to a thesis if price disconnects).

By July 28, 2026, the PHLX Semiconductor Index (SOX) had fallen 24.5% from its June peak — formally in bear-market territory — and the Nasdaq-100 was down 9.7% from its record, sitting on the edge of correction (MarketWatch, NBC News). Triggers stacked: a Broadcom fiscal Q2 print on June 3 that beat estimates but failed to extend the upside gap; reports of progress in Chinese advanced chipmaking; growing scrutiny of AI circular financing (Nvidia down nearly 5% in a single session per Yahoo); and TSMC/Netflix earnings that disappointed the bid for capex durability.

This is the first time since 2022 that a coordinated, broad-based chip selloff has coincided with a high-profile thematic-fund liquidation. The previous AI drawdowns (summer 2024, spring 2025) were absorbed by ETF and fundamental flows; this one runs through prime-broker financing lines.

Upstream of the deal are the four prime brokers — Goldman Sachs, JPMorgan Chase, Bank of America and Citigroup — that intermediated the block sale. Their prime brokerage books are the financing pipes that powered Situational Awareness' leverage. Per Reuters (Jan 2026), all four reported double-digit prime-brokerage revenue gains through 2025, driven by exactly this kind of hedge-fund leverage. They didn't take the loss (Citadel did), but they also didn't lose the customer — the brokers continue to clear the surviving $10B book and the rebuilt fund.

Downstream, the named public-equity victims are the same names Situational Awareness held. The most leveraged of these structurally — CoreWeave — has the weakest fundamentals to absorb a marginal-long disappearance. Per FMP, CoreWeave trades at 11.4x EV/Sales with a -27% free-cash-flow yield, debt/equity of 7.4x, and interest coverage of -0.38x. Its float is dominated by insiders (18.3%) and institutions (58.5%). When the only thematic funds owning this name sell into Citadel's bid, the secondary market has fewer natural buyers.

Aschenbrenner's decision to keep the Anthropic stake and exit the public-equity book tells you where the smart money thinks the next dollar of conviction lives. Public AI names are now valued by a market-maker's inventory model, not a thematic-fund model. Private AI labs — where capital is locked up, where there is no daily mark-to-market, and where the only price is the last primary round — sit at the other end of the cycle. Per Cracking the Vault, Anthropic had raised $65B at a $965B valuation. Situational Awareness kept its slice rather than mark it down.

On the short-term horizon (days to quarters), the trade is mechanical. Citadel will likely distribute a portion of the block to its own pods and to other multi-strategy buyers over weeks, providing a soft landing. Any relief rally in chip names will be capped by Citadel's willingness to sell into strength. The Nasdaq-100 remains within 1% of correction; a 10% drop from peak is the line where forced-buyer demand historically accelerates.

On the long-term horizon (1–3 years), the structural takeaway is that AI-chip valuations will price more like other late-cycle commodity-correlated industries — bid-supported, thesis-light — and less like the 2023–2025 thematic regime. The new buyer of last resort is a $67B multi-strategy shop with a P&L-driven exit, not an AGI believer with a 13F. That compresses upside skew on the entire chip complex and re-rates the listed AI-trade toward cash-flow durability over thematic conviction.

Citadel's absorption of Situational Awareness' $16B book replaces the marginal thematic long with a market-maker's inventory, leaving Broadcom, Nvidia, AMD and CoreWeave with firmer floors but capped rebounds — and signals that, in this AI correction, the only conviction left sits in private AI labs like Anthropic, where there is no price to mark down.