Citadel Bought $16 Billion in AI Stocks From Situational Awareness After Leverage Turned a 439% Gain Into a Margin Crisis

Ken Griffin’s Citadel has acquired the bulk of Leopold Aschenbrenner’s AI stock portfolio after Situational Awareness, the hedge fund built on an AGI investment thesis, was forced to sell roughly $16 billion in public equities during July’s AI market correction. The Wall Street Journal first reported the transaction on July 30. The Financial Times had previously reviewed Aschenbrenner’s July 24 investor letter, which disclosed a 439% net return for H1 2026.

The deal was assembled in under 24 hours with four prime brokers facilitating the transfer: Goldman Sachs, JPMorgan Chase, Bank of America, and Citigroup, according to Disruption Banking. Bloomberg reported that Millennium Management and Jane Street examined the portfolio but did not participate. Jane Street is itself an investor in Situational Awareness, a fact Disruption Banking noted is unusual given how rarely the firm backs external managers.

From Essay to $20 Billion Fund

Aschenbrenner published “Situational Awareness: The Decade Ahead” on June 4, 2024, via X. The essay argued that continued scaling of compute and algorithmic efficiency made AGI by approximately 2027 plausible, with automated AI research potentially triggering a rapid intelligence explosion. That thesis became the investment framework for Situational Awareness LP, launched in late 2024 after Aschenbrenner was fired from OpenAI’s Superalignment team in April of that year, as reported by Disruption Banking.

The fund attracted early capital from Stripe co-founders Patrick and John Collison, former GitHub CEO Nat Friedman, and investor Daniel Gross. Jane Street joined later. Within two years, Aschenbrenner, who graduated from Columbia as valedictorian at 19, had scaled the fund to roughly $20 billion in assets under management with almost no prior professional investing experience, per Disruption Banking.

The H1 Strategy: Infrastructure Long, Disrupted Software Short

The 439% H1 net return came from a directional bet on AI infrastructure buildout. Public holdings included Riot Platforms, Core Scientific, CleanSpark, Bloom Energy, AMD, and Oracle, according to Disruption Banking. The fund’s private investments, per Dealroom data cited in the same report, included positions in Anthropic, Fluidstack, MatX, T1 Energy, and Sharon AI.

The thesis was not simply “AI stocks go up.” It was a structural view: the path to AGI demands far more chips, memory, and power than markets have priced in, while existing software companies that rely on manual workflows face disruption from autonomous agents. Long the picks and shovels, short the incumbents.

That view was correct in H1 2026. Nvidia’s data center revenue continued to beat estimates. ASML and TSMC maintained pricing power. AI infrastructure stocks outperformed the broader market. The fund’s leverage, which the Financial Times reported was used to magnify returns, amplified every upward move.

July: When Leverage Meets Volatility

The same leverage that produced 439% in six months created a liquidity crisis in July. Goldman Sachs estimated that Asia-focused fundamental long-short funds lost an average of 18.6% through July 28, according to Disruption Banking. AI chip and infrastructure stocks, which had been Situational Awareness’s core positions, sold off alongside the broader rout.

People familiar with the matter told the Wall Street Journal that the fund needed cash to meet margin-call demands from lenders. Reuters was more cautious, reporting that it was unclear whether lenders had formally issued margin calls before the deal was struck. Either way, the fund faced a binary choice: raise fresh capital from existing investors or offload the leveraged book. It chose the portfolio sale.

In his July 24 investor letter, written before the Citadel deal, Aschenbrenner had framed the selloff as creating “some of the most attractive opportunities since early 2025” and invited investors to commit additional capital starting August 1, with a potential Anthropic IPO as a near-term catalyst, per Disruption Banking. Less than a week later, Citadel was buying and Situational Awareness was selling.

The Citadel Playbook

Griffin’s firm has done this before. In 2006, Citadel and JPMorgan absorbed Amaranth Advisors’ energy portfolio after its natural gas positions collapsed. In 2007, Citadel took on Sowood Capital’s book. The pattern is consistent: a single buyer removes a forced seller from the market instead of letting positions bleed out over weeks through broker-facilitated block sales, as reported by Disruption Banking.

That is also what distinguishes this unwind from Archegos in 2021, when banks sold blocks in competition with one another and several took heavy losses.

The timing has drawn scrutiny. On July 27, Frank Flight, head of macro strategy at Citadel Securities, argued publicly that the Federal Reserve could deliver a surprise quarter-point rate rise, calling it a potential “cleansing event” for market pricing. The Fed instead held its target range at 3.5% to 3.75% on July 29, though three of twelve voting members preferred an increase.

Shay Boloor, chief market strategist at Futurum Equities, posted on X that Citadel had bought Situational Awareness’s “assets at significantly lower prices knowing that if sentiment deteriorated enough the fund could be forced to sell,” calling the sequence “absolutely ruthless.” But as Disruption Banking noted, Citadel Securities (the market maker) and Citadel (the hedge fund) are separate entities, and none of the WSJ, FT, or Reuters reports linked the rate-hike note to the portfolio acquisition.

Several stocks associated with Situational Awareness rebounded sharply on July 30 after the transaction removed the threat of a disorderly market sale.

What Survives

Situational Awareness has not closed. Reuters reported the fund retains a book of roughly $10 billion, comprising remaining stock positions and private investments. The Anthropic stake was not sold.

That Anthropic position may now be the fund’s most consequential asset. Anthropic confidentially submitted its draft S-1 on June 1, 2026, putting an IPO on a near-term timeline. If Anthropic prices at or above its last private valuation, the return on that single position could offset meaningful losses elsewhere. Aschenbrenner highlighted the potential IPO as a catalyst in his July 24 letter, according to the Financial Times.

The Leverage Lesson for AI Capital Markets

Aschenbrenner’s core thesis remains intact. The positions he selected — AI infrastructure providers and energy companies serving data center expansion — remain central to the buildout narrative that Nvidia, OpenAI, Microsoft, and Google continue to reinforce through their capital expenditure commitments. What the episode exposed was a risk management failure: a correct thesis applied at leverage levels that could not survive a one-month correction.

A 439% six-month return on a $20 billion book implies leverage in the range of 3-4x, consistent with what the Financial Times described as “borrowing to magnify returns.” At that level of leverage, a 15-20% drawdown in the underlying positions can erase the entire equity cushion and trigger forced liquidation, regardless of whether the long-term thesis is correct.

That dynamic matters beyond one fund. The AI infrastructure trade has attracted concentrated capital across hedge funds, sovereign wealth funds, and pension allocators. If Situational Awareness, with its deep thesis conviction and high-profile investor base, could not hold through a one-month correction, other levered positions in the same trade are similarly vulnerable to forced selling during future drawdowns.

The broader capex cycle has not slowed. OpenAI is in advanced talks for a Nvidia-backed $250 billion financing backstop for its Pike County data center campus. Microsoft reported continued acceleration in its AI infrastructure spend during Q4 FY2026 earnings. But the gap between “the thesis is right” and “the trade works on a levered 30-day timeline” is exactly where funds blow up.

Aschenbrenner argued in 2024 that the path to AGI would demand far more infrastructure than markets expected. Nothing in July’s selloff has settled that argument. Whether the thesis survives is now a separate question from whether the fund that carried it does.