Last week, the sell-off in global semiconductor stocks, closely tied to the AI computing infrastructure theme, reached its most intense point of the year. The youngest hedge fund manager, hailed as an 'AI investment prophet' for his forward-looking industry vision, was also caught in the storm. As highly leveraged AI bets collapsed, Leopold Aschenbrenner's hedge fund, Situational Awareness, suffered a record 67% loss in July, forcing it to sell most of its public market holdings to Citadel, the hedge fund giant led by Wall Street billionaire Ken Griffin, and completely unwind all leverage.
Leopold Aschenbrenner's rise to fame came from translating a highly insightful technical perspective directly into capital market positions. Graduating first in his class from Columbia University at 19, he joined OpenAI's 'Superalignment' team. In 2024, he published a 165-page paper, 'Situational Awareness,' predicting that AGI would force global exponential demand for a complete AI data center delivery chain, encompassing data center power equipment, liquid cooling, server CPUs, DRAM/NAND/HBM, optical communication/interconnects, high-performance Ethernet network infrastructure, transformers, gas turbines, and energy storage systems. He subsequently founded the eponymous fund, attracting capital from notable investors like Jane Street and Stripe's founders, with its assets under management exceeding $20 billion within two years. By the end of June 2026, the fund's year-to-date net return was an astonishing 439%, and its total return since inception surpassed 1000%, making it the most legendary concentrated bet of the current AI bull market, earning him praise from some retail investors as the 'AI investment prophet' and the 'ultimate answer of the AI super-investment boom.'
After the 'AI prophet' fell, some Wall Street analysts believe that with short covering, institutional and retail buying on dips, and the prospect of continued strong expansion in AI computing fundamentals, the de-leveraging process in global AI tech stocks is nearing its end. This set the stage for a historic single-day super rebound in the US semiconductor sector last Thursday, bringing back the familiar feeling of the AI computing frenzy. In both the US market last Thursday and the Asia-Pacific market on Friday, the semiconductor sector, closely tied to AI computing infrastructure, experienced a massive rebound shifting from 'forced liquidation of extreme leveraged positions' to 'vengeful short covering.' When examining the recently released stellar earnings and outlooks from AI computing chain leaders like Lam Research, Samsung Electronics, UMC, Taiwan Semiconductor Manufacturing, SK Hynix, and Seagate Technology, combined with South Korea's 179% year-over-year surge in July semiconductor exports, the conclusion seems clearer: the relentless physical demand for AI computing infrastructure has not deteriorated in sync with the stock price and extreme leveraged position sell-off and liquidation.
The viewpoint was forward-looking, but the position collapsed first! The 'AI prophet' faced Wall Street's oldest liquidity trial. 'This month we let you down,' Leopold Aschenbrenner wrote in a letter to investors. However, even after the remarkable pre-sell-off rally, the Situational Awareness fund was still up about 80% year-to-date. What collapsed first wasn't the investment thesis, but the financing structure. This moment carries a somewhat unsettling echo of Wall Street history. Two tech booms followed similar trajectories in their first four years – even triggering a hedge fund leverage crisis at almost the same point. Bespoke Investment Group charted the performance of the Nasdaq index after the launch of Netscape in 1994 and ChatGPT's global debut in 2022. The two tech booms showed remarkably similar paths in their early years, and the liquidation of Situational Awareness's positions occurred around the same time as Long-Term Capital Management (LTCM) nearly collapsed in 1998.
The key difference is that Situational Awareness never posed a systemic threat anywhere near the scale of LTCM. The similarities are more fundamental. A brilliant AI genius discovered a powerful idea. They concentrated their bets on the AI computing theme. Then, leverage made time the master. Aschenbrenner saw early on the potential scale of the AI data center construction wave. His 2024 forward-looking white paper, 'Situational Awareness: The Decade Ahead,' depicted potential exponential growth in global computing power and extrapolated deeply along these trends. 'Large US companies are preparing to invest trillions of dollars, unleashing an industrial mobilization of American power not seen in a long time,' he wrote. As new data centers break ground, power agreements are signed, and overall capital expenditure plans for AI computing infrastructure hit new highs, this prediction no longer seems radical. The AI investment boom is already driving corporate cash flow towards AI chips, server clusters, high-performance network equipment, and data center power equipment. But an investment cycle measured in years has collided with a leverage financing structure that can be tested in days.
Asset price declines can trigger margin calls, forcing investors to sell assets before their investment thesis is validated. Leverage allows investors to borrow money to control a larger portfolio. When asset prices rise, borrowed money amplifies gains. This mechanism works just as quickly in reverse. Declining position prices reduce the value of assets serving as loan collateral. Lenders may demand more cash or a smaller portfolio. This forces investors to quickly sell liquid assets, regardless of whether the initial investment thesis has changed. Situational Awareness stated that the rapid decline in position prices and decreasing market liquidity made it increasingly difficult to maintain the portfolio within risk limits. Aschenbrenner compared this dynamic to a bank run – every sign of vulnerability creates more vulnerability. The hedge fund eventually sold most of its holdings in its public stock portfolio and completely eliminated leverage. It retained its private market investments, including its stake in Anthropic. The portfolio likely sold not Aschenbrenner's least favored assets, but those easiest to liquidate quickly. Public market stocks trade all day with constantly changing market prices. Private holdings may be harder to sell, but they also don't provide lenders with such an immediate liquidation channel. When pressure rises, liquidity can determine which assets are retained and which must be sold. Wall Street has seen this movie before, only the starring assets keep changing.
A brief history of classic Wall Street leverage blow-ups is shown below. Concentrated holdings decline. Borrowed money magnifies losses. Margin calls follow, and investors lose control over the exit timing. Different markets, same trap. This explains why the brutal cost of leverage extends far beyond hedge funds. Leveraged ETFs, margin accounts, or options positions can all create the same mismatch between an investor's time horizon and the market's timeline. Aschenbrenner's AI investment thesis may ultimately prove correct. Removing leverage buys the fund more time to validate it. But Wall Street won't reward you with more time just because you are smart. Long-term investment theses can wait, but lenders won't. The most profound conclusion from this drama is that Aschenbrenner may have correctly identified the long-term direction of the AI computing industry but mistakenly equated technical certainty with stock market path certainty. Demand for HBM, NAND, advanced packaging, CPUs, optical interconnect equipment, power, and liquid cooling may still grow, but a successful investment strategy must withstand valuation compression, correlation shifts, tighter financing conditions, and crowded trade reversals. The core of the 'Kelly Criterion' is not 'bet heavy because the odds are high,' but to maximize long-term compounding and avoid ruin; once a position exceeds the optimal size, even if the industry judgment is ultimately correct, short-term variance can be enough to liquidate the investor before the theme plays out.
Situational Awareness proved the cruelest rule of this AI bull market: the market doesn't reward the one who sees the future first, but the one who both sees the future and has enough liquidity to survive until it arrives. 'This is indeed part of the selling pressure that has persisted over the past few weeks,' said Calvin Yeoh, co-manager of the Merlion Fund at Blue Edge Advisors, commenting on the hedge fund's ordeal. 'At the same time, almost all retail investors in South Korea using extreme leverage strategies have been wiped out, leaving no more significant sellers there.'
After the 'AI prophet' fell, has the familiar market feeling of the AI computing frenzy finally returned? Citadel's assumption of most of Situational Awareness's public stock positions, combined with record foreign capital inflows into South Korea, stricter regulation of leveraged ETFs, and surging semiconductor exports, largely suggests that the most dangerous 'forced deleveraging – margin call – indiscriminate selling' negative feedback loop in the AI computing chain has significantly eased. However, it more accurately marks the beginning of a fundamentally supported strong recovery rally, not the confirmation of a new wave of indiscriminate semiconductor bull market frenzy. In other words, the end of deleveraging helps form a market bottom but does not mean a new leverage boom has started. To truly confirm the start of a semiconductor bull run, major semiconductor benchmark indices must not break key technical support levels on pullbacks, net inflows and earnings upgrades must continue in the coming weeks, leveraged ETF funds in the Korean and US stock markets must not see violent outflows, and cloud giants like Microsoft and Amazon.com must continue to prove that capital expenditure can translate into cloud revenue, order backlogs, and acceptable asset payback periods.
Situational Awareness, due to its concentrated, leveraged bet on AI stocks, faced margin calls and was forced to transfer most of its public stock portfolio to Citadel. Subsequently, Wall Street traders partly attributed the July 31 tech stock rebound to the removal of this major selling pressure source. The market microstructure significance is clear: a large, price-insensitive seller exits, margin call pressure from prime brokers stops, other investors no longer need to sell ahead of it, and short covering and market maker hedging amplify the rebound. The strong fundamental outlook for the AI computing infrastructure industry is providing more durable fuel for this semiconductor rebound than short covering alone. The most compelling fundamental evidence for this rebound comes from South Korea's July exports. Total exports surged 62.8% year-over-year to $98.89 billion, beating expectations. Semiconductor exports jumped 179%, and computer exports rose 404%, driven by AI data center investment, rising memory chip prices, and SSD demand. Exports not only continued to grow strongly but also saw significant increases in shipments to both the US and China, indicating that global AI hardware demand has not weakened in tandem with July's stock price crash.
Microsoft's latest quarterly Azure revenue grew 43%, beating the ~39.98% market estimate, suggesting AI capital spending is starting to translate into cloud revenue and cash flow. Amazon.com's AWS revenue grew 37% to $42.2 billion, significantly exceeding expectations, with its order backlog jumping from $364 billion to $496 billion. Management stated that most of its computing capacity for 2027 is already booked by customers, and even with a 2026 capital expenditure increase to $220 billion, it cannot fully meet demand. Lam Research's next-quarter revenue guidance midpoint reached $8.1 billion, well above the market consensus of $7.09 billion. These recent industry data points prove that AI demand is spreading from GPUs to HBM, DRAM, NAND, advanced packaging, etching/deposition equipment, and enterprise SSDs. From the perspective of major Wall Street banks, as this round of forced deleveraging liquidation concludes, the new AI computing super bull market shows 'signs of re-ignition,' but it is not yet possible to declare the main uptrend fully confirmed from a capital flow perspective. Situational Awareness clearing its leverage proves that the most violent mechanical selling pressure may be past, but it does not prove that all over-leveraged institutions have been cleared out. Even with a near 18% single-day surge, the KOSPI is still about 30% below its all-time high. The Philadelphia Semiconductor Index was still down over 20% for July, having previously corrected nearly 30% from its June high.
The familiar, vibrant, bullish fundamental sentiment for AI computing may be returning, but the main uptrend trajectory of the AI-driven super bull market still needs continuous confirmation from earnings, orders, AI capital expenditure returns, and the ability to not break technical lows on pullbacks. JPMorgan states that the real support for the medium-to-long-term bullish thesis of the Korean stock market and AI computing-related semiconductor stocks is the near-insatiable demand for computing infrastructure like HBM, server DRAM, enterprise SSDs, and advanced memory from AI data centers, coupled with supply discipline supporting spot and contract prices. With foreign capital returning to the Korean stock market last week, hitting a record single-day inflow, Morgan Stanley's publicly released report, 'Buying the AI Infrastructure Dip,' attributes the recent decline in AI infrastructure stocks primarily to short-term positioning, crowded trades, and technical factors, rather than deteriorating demand fundamentals. Its core judgment is that AI computing demand will likely far exceed supply for years to come. More efficient, lower-cost models could paradoxically expand token usage and total computing demand through the Jevons paradox. Constraints like power, skilled labor, and permitting are 'speed bumps' slowing construction, not 'brick walls' stopping industry expansion. Morgan Stanley estimates that total capital expenditure for the top five US tech giants could rise from nearly $800 billion in 2026 to approximately $1.2 trillion in 2027 and $1.4 trillion in 2028. This provides extremely high revenue visibility for AI computing infrastructure assets like GPUs, HBM, Ethernet switch chips, optical communication/interconnects, cooling infrastructure, data center CPUs, power equipment, and powered shells. Simultaneously, such massive capital expenditure means the market's next phase will not just reward 'getting orders and outlooks' but will strictly scrutinize AI revenue increments, free cash flow, cost per token, asset depreciation life, and financing costs. In Morgan Stanley's view, the long-term AI computing infrastructure bull market is not over, but the best opportunities after the correction lie in companies that genuinely control bottlenecks in computing, storage, power, and data center delivery, not in all high-beta assets labeled with AI.
Comments