16 Billion "Blood-Stained Chips" Settled: Citadel Defuses the AI Crash Wave—Is This the End of Liquidation or the Start of a Domino Effect?

According to Zhitong Finance APP, Citadel, the hedge fund giant led by Wall Street billionaire Ken Griffin, has reached a deal with Situational Awareness, an up-and-coming hedge fund mired in collapse, helping to trigger a global surge in AI computing power-themed stocks—especially semiconductor stocks closely tied to AI computing power. This has ignited a dramatic “irrational bullish sentiment” super rally, prompting heated debate among Wall Street traders: has the worst stage of AI-driven selloffs ended, or are there still more stock market blowups ahead within the AI infrastructure supply chain?

On Thursday in the US stock market and Friday across Asia-Pacific equities, semiconductor sectors closely connected to AI infrastructure saw a spectacular rebound from “forced liquidation due to extreme leveraged positions” to “aggressive short covering.” When reviewing recent performance and outlooks announced by AI computing leaders such as Lam Research, Samsung Electronics, UMC, TSMC, SK Hynix, and Seagate, the conclusion seems clearer: the physical layer demand for AI-related infrastructure hasn’t deteriorated in tandem with plunging stock prices and deleveraging positions.

After three consecutive trading days of circuit-breaker-level plunges, the Korea Composite Stock Price Index (KOSPI), home to super memory chip giants SK Hynix and Samsung Electronics, soared a record 18%. Taiwan’s benchmark stock index, where TSMC—the “king of foundries” controlling nearly all global AI chip production—carries massive market cap weight, rose 8%. Japan’s Nikkei 225 Index, boasting leading AI infrastructure names such as Kioxia, Tokyo Electron, and Advantest, also rose 4%. On Thursday, the closely watched US semiconductor index (the Philadelphia Semiconductor Index—a “global semiconductor bellwether”) posted its biggest gain since April 2025.

As shown above, the Korea Composite Stock Price Index skyrocketed about 18%, signaling a full return of AI optimism.

For some market participants, Situational Awareness being forced to liquidate nearly all its public equity AI positions helps explain the relentless global semiconductor sector deleveraging and aggressive selling seen throughout July: a single wave of selling reverberates across markets and can morph into new rounds of super selloffs. Before retail and professional investors made substantial cuts to borrowing, margin financing in Korea’s financial markets hit an all-time high last month.

After Citadel acquired most public stock holdings at a discount, the market was relieved of high-leverage hedge funds dumping shares across exchanges tick by tick. Other investors no longer worried about massive unknown forced liquidations. This, combined with short covering, dealer hedging adjustments, and fundamental flows, jointly fueled Thursday’s violent semiconductor stock rebound globally.

Citadel taking over the majority of Situational Awareness’s public portfolio amounted to removing a steady, indiscriminate, price-insensitive forced seller from the market. Together with technology titans like Microsoft and Amazon easing concerns over AI investment returns, plus TSMC, UMC, and Seagate sending highly bullish signals regarding AI computing supply, KOSPI’s record 17.91% surge followed, accompanied by a revenge rally in global chip stocks led by Micron, AMD, TSMC, Samsung Electronics, and SK Hynix. This fundamentally means that marginal price setters in the market have quickly shifted away from margin calls, leveraged ETF redemptions, and forced liquidations, moving instead toward short covering, institutional and retail dip buying, and optimism around robust AI infrastructure fundamentals.

While Situational Awareness’s post-leverage cleanout shows the most intense mechanical selling may have passed, it doesn’t prove all highly leveraged players are gone; even with a near 18% daily jump, KOSPI remains almost 30% below its historical high; the Philadelphia Semiconductor Index is still down over 20% for July, having at one point retreated nearly 30% from its June peak. The vibrant, exuberant, AI-driven risk-on mood may be back, but confirmation of a true super bull run in AI will require earnings, orders, AI capital expenditure returns, and technical retests that do not break major support levels.

From a 165-page AGI prediction to “the most crowded long trade” blowing up: the ‘AI version answer’ got the sector right but lost the trade, a 439% myth defeated by the leverage cliff

Leopold Aschenbrenner’s brilliance lay in translating piercing technical insights directly into capital market positioning: at 19, he graduated first in his class from Columbia University, joined OpenAI’s “Superalignment” team, and published a 165-page “Situational Awareness” in 2024. He predicted AGI would force the world to make exponential purchases across the complete AI data center supply chain—data center power equipment, liquid cooling, CPUs, DRAM/NAND/HBM, optical interconnects, high-performance Ethernet, transformers, gas turbines, and so on.

He then founded a fund of the same name, attracting major investors like Jane Street and Stripe’s founder, growing assets under management to over $20 billion in two years. By end-June 2026, the fund’s annual net return hit an astonishing 439%, with over 1000% cumulative returns, making it the most legendary concentrated AI bull play of this cycle—revered by some retail investors as “the version answer in the AI super investment boom.”

His undoing was not a sudden invalidation of the long-term AI thesis, but the same macro narrative controlling both longs, shorts, and the financing leg. Situational Awareness crowded into AI storage, cloud compute, and power infrastructure assets such as SK Hynix, SanDisk, Nebius, CoreWeave, Bloom Energy, while shorting some software companies seen as AI-disrupted. When the AI hardware crowded trade reversed in July, longs dropped together and some shorts rallied, making correlations approach 1 in stressed times—turning the seemingly “long-short hedged” portfolio into one losing in all directions.

The fund plunged about 67% in July, forcing a choice between “rapidly raising new capital” or “selling the public portfolio.” Most of the $16 billion public stock portfolio was ultimately transferred to Citadel. Prime brokers Goldman Sachs, JPMorgan, Bank of America, and Citi coordinated the deal, leaving the fund with about $10 billion predominantly in private equity (including Anthropic), so headlines like “fund wiped out” or “Citadel acquired all assets” aren’t accurate—the real destruction was its high-leverage public strategy.

The so-called “overnight collapse” was actually weeks of mark-to-market losses crossing a financing cliff within 24–36 hours, morphing into a liquidity event. The danger in leverage isn’t getting the final direction wrong—the net asset value (NAV) is just a thin layer under a huge gross exposure: at 4x gross leverage, a 25% portfolio drop theoretically wipes out all the equity; brokers then raise margin requirements, cut credit lines, and force selling. Once the market sniffs out that an institution has become a price-insensitive forced seller, trading desks front-run the unwind, worsening execution prices further.

Citadel’s key role wasn’t “saving a genius” but converting weeks of continuous forced selling across exchange orderbooks into a single block asset transfer to a deeper-pocketed institution. Once the market saw Situational Awareness didn’t have to keep dumping, seller supply dried up, and short covering plus fundamental buyers stepped in: KOSPI surged a record 17.9%, Samsung Electronics and SK Hynix soared about 28% and 30%, and the Philadelphia Semiconductor Index strongly rebounded.

Is the AI crowded trade blowup truly over, or just the eve of dominoes falling?

The most profound conclusion from this episode: Aschenbrenner may have judged the long-term direction for AI infrastructure rightly, but wrongly equated technological certainty with stock price trajectory certainty. HBM, NAND, advanced packaging, CPUs, optical interconnects, power and liquid cooling demand may continue rising, but great investment strategy must survive valuation compression, correlation spikes, tightening financing, and crowded position reversals. The “Kelly Criterion” isn’t “heavy up the highest probability bet,” but maximizing long-term compounding while avoiding ruin; exceed the optimal risk size and, even with correct industry themes, volatility can force exit before the thesis pays off.

Situational Awareness embodies the current AI bull market’s hardest rule: the market rewards not the ones who first see the future, but those who both see it and have enough liquidity to survive until it becomes reality.

“That was definitely part of the persistent selling pressure over the past few weeks,” said Calvin Yeoh, co-manager of the Merlion Fund at Blue Edge Advisors, commenting on Leopold Aschenbrenner’s fund. “Meanwhile, all Korean retail traders using extreme leverage have essentially been wiped out—there aren’t bigger or more sellers now.”

“But that doesn’t mean the market will just rally all the way back,” Yeoh added. “It just means the levered unwinds are mostly over for now.”

The rebound of the past two days closes a turbulent July. The Philadelphia Semiconductor Index remains down over 20% this month, its worst monthly showing since the global financial crisis.

KOSPI is down 22% in July, with SK Hynix and Samsung Electronics the main drags—even though these stocks led the index to world-beating returns earlier this year. SK Hynix was one of Situational Awareness’s holdings, and its US ADR briefly fell below its issue price.

The dramatic implosion of Situational Awareness immediately recalls Long-Term Capital Management (LTCM) and 2008.

Yeoh is not alone in viewing the selling pressure as temporarily eased, but this event has raised doubts among investors whether other funds are facing similar trouble. Some on trading desks compared it to the swift collapse of LTCM in 1998, when US Federal Reserve policymakers organized a rescue due to the event’s systemic risk for financial markets.

Other analysts were reminded of the global financial crisis days, when banks and funds were repeatedly taken over.

“If there’s already one Situational Awareness, how many more similar institutions are out there?” said Cusson Leung, chief investment officer for KGI International Wealth Management. “This morning, it really reminded me of JP Morgan’s purchase of Bear Stearns in 2008,” he said in an interview. “At that time, there was a relief rally too, but then Lehman Brothers still went bankrupt.”

Situational Awareness’s asset value plummeted from $45 billion at the start of July to about $10 billion. Aschenbrenner had to liquidate positions to meet margin calls. Then, according to a person familiar with the matter who requested anonymity due to private information, Griffin’s Citadel reached out and acquired most of Situational Awareness’s investment assets at a discount.

“The bigger question is whether this is the only overleveraged sector in the market,” said Ulrich Urbahn, head of multi-asset strategy and research at asset management giant Berenberg. “If other funds are similarly concentrated in AI, compute, and semiconductors, this could be just the first domino and not the last.”

Exclusive data from Bloomberg Intelligence shows that over the past two weeks, leveraged ETFs linked to Samsung Electronics and SK Hynix (single-stock leveraged ETFs) have seen massive net outflows; previously, these were seen as amplifying market volatility in Korea. As of Thursday, Bloomberg Intelligence data showed assets under management in these ETFs had plunged from more than $11 billion on June 25 to $4.1 billion. Korea’s government pledged more measures this week to stabilize markets and said it would restrict retail investor participation in leveraged ETFs.

“We’ve seen big leveraged bets being unwound, and that process is nearing completion. The rescue of Situational Awareness was the last missing piece, at least in the short term,” said Joshua Crabb, head of Asia-Pacific equities at Robeco. “Asia benefited sharply because valuations were already low and some AI infrastructure-related tech names in Korea had become extremely cheap.”

Currently, this rebound is driven by lower positioning, improved valuations, and better sentiment.

“This episode delivers the same timeless lesson as LTCM,” said Amir Anvarzadeh, equity strategist at Asymmetric Advisors in Japan. “As Keynes famously put it: ‘The market can stay irrational longer than you can stay solvent.’”

After the fall of the AI oracle, is that familiar AI computing power euphoria finally back?

Margin rules have effectively flushed out highly leveraged buyers unable to continue financing. When funds crowd into SK Hynix, storage, compute infrastructure, and other AI assets, expanding exposure with borrowings, falling stocks reduce collateral value, lenders then demand more cash or collateral; funds trying to meet margin calls must keep selling the same assets, creating a “drop—margin call—sell—further drop” negative feedback loop. With Citadel scooping up most public holdings at a discount, the market no longer suffers piecemeal forced selling from this fund, nor do others fear huge, unknown liquidations, allowing short covering, dealer hedging, and fundamental flows to together fuel violent rebounds.

The core strategy of Situational Awareness was essentially to use high leverage to express a judgment that may be right long-term but extremely risky short-term: AGI and intelligent agents will exponentially expand global compute demand, so concentrate on semiconductors, storage, data centers, power, and private AI ventures, magnifying returns with leverage.

Public news confirms Situational Awareness used substantial leverage and was hit by margin calls from lenders, but detailed contractual composition is yet undisclosed—there’s no definitive proof all positions were built via total return swaps (TRS). Typically, with TRS, a fund posts some initial margin to a prime broker to receive payout from a far larger basket of stocks; the fund gets the upside minus interest, but absorbs all downside losses. The prime broker hedges by buying the stocks; if prices fall and the fund can’t post variable margin, the bank terminates the swap, sells the hedge or collateral, converting account losses into real market selling.

For Wall Street’s bulge-bracket banks, the current round of forced deleveraging seems to be ending and signs of a new AI compute super bull market “restart” are emerging, though a confirmed, well-funded leg up isn’t certain yet.

J.P. Morgan strategist Mixo Das recently forecast that the unwinding process in Korea’s leveraged ETFs is likely done, with hedge fund deleveraging around 90% complete, meaning “irrational sellers” driving falling knives have almost exhausted themselves. This corroborates the Citadel–Situational Awareness deal in which some $16 billion of equities were transferred—turning unrelenting forced selling into a single block, allowing the downward spiral in global AI assets to stop.

J.P. Morgan’s research shows that Korean leveraged ETF assets fell from a $50 billion peak to $17 billion, with hedge fund long-short leverage ratios from 5.7x to 3.2x; these numbers mean feedback loops of forced selling from leveraged ETFs, swap financing cuts, and momentum strategies are waning.

Valuations in Korea’s stock market have indeed hit crisis-level attractiveness, but a forward P/E of 5 cannot, on its own, confirm an absolute bottom. The memory industry is highly cyclical and when profits are high, P/Es can look artificially low—sometimes the “cheapest P/E” historically coincides with peak earnings.

Fundamentals, meanwhile, are now providing propulsion for the semiconductor rebound that extends beyond mere short covering. Microsoft’s latest quarterly Azure revenue soared 43%, beating the market’s 39.98% forecast and showing AI CapEx translates to cloud income and cash flows. Amazon AWS revenue rose 37% to $42.2 billion, far above expectations, with backlogged orders jumping from $364 billion to $496 billion, and management said most compute capacity is already booked through 2027—even raising 2026 CapEx to $220 billion falls short of demand. Lam Research’s next-quarter revenue guide midpoint is $8.1 billion, well above Street consensus ($7.09 billion), proving AI demand is spreading across GPUs to HBM, DRAM, NAND, advanced packaging, etch/deposition, and enterprise SSD.

To truly confirm the main uptrend in semiconductors, all the major sector bellwethers must not break key technical levels during pullbacks, Korean leveraged ETF outflows must stop, and cloud giants like Microsoft and Amazon must keep proving CapEx translates to cloud revenue, backlogs, and palatable asset ROIs.

J.P. Morgan notes that the mid-to-long-term bull logic for Korean stocks and AI compute-related semiconductors rests on relentless expansion of compute infrastructure in AI data centers—demand for HBM, server DRAM, enterprise SSDs, and advanced memory. Supply discipline means spot and contract prices remain supported. Strong earnings from Amazon, Google, and Microsoft have assuaged concerns about slowing AI CapEx, propelling KOSPI to a record 17.91% jump on July 31, with Samsung Electronics and SK Hynix up about 28% and 30% respectively. This shows the rebound is driven not only by short covering but also by renewed conviction that “AI CapEx will likely continue transforming into cloud revenue, storage orders, and chip profits”—a bullish fundamental reappraisal for AI.