Citadel takes away the "AI Prophet" leverage bomb, the most dangerous selling pressure on the computing power chain is receding! Semiconductors transition from leveraged strangulation to "fundamental prosperity pricing."

Last week, a wave of sell-offs in semiconductor stocks closely linked to AI computing infrastructure reached its peak for the year in global equity markets. The youngest hedge fund manager, nicknamed the "AI investment oracle," recognized for his forward-looking view on the industrial chain during the AI boom, also found himself at the eye of the storm. As high-leverage bets on AI collapsed, the hedge fund Situational Awareness, led and managed throughout by Leopold Aschenbrenner, recorded a historic loss of 67% in July, compelling the fund to sell most of its public market holdings to billionaire Ken Griffin's hedge fund giant Citadel and to completely unwind all leverage. Leopold Aschenbrenner's brilliance stemmed from his ability to directly translate an incisive technical judgment into capital market positions: he graduated first in his class from Columbia University at 19 and joined the OpenAI "super-alignment" team. In 2024, he published a 165-page report titled "Situational Awareness," proposing that AGI would force the world to acquire AI data centers covering the entire supply chain, including power equipment for data centers, liquid cooling, CPUs, DRAM/NAND/HBM, optical communication/optical interconnects, high-performance Ethernet network infrastructure, transformers, gas turbines, and data center energy storage systems at an exponential demand pace. Subsequently, he established his eponymous fund, attracting renowned capitals such as Jane Street and Stripe's founders, with assets under management exceeding $20 billion within two years. By June 30, 2026, the fund achieved an astonishing net return of 439% for the year and over 1000% since its inception, becoming the most legendary concentrated bettor in this round of the AI bull market, with some retail investors even hailing him as the "AI investment oracle" and "the version answer under the AI super investment craze." After the fall of the AI oracle, some Wall Street analysts believe that with short covering, institutional and retail buying at lower prices, and strong fundamentals for AI computing continuing to expand, the global de-leveraging process in AI tech stocks is nearing its end. Last Thursday, the U.S. semiconductor sector staged a historic single-day super rebound, and the familiar feeling of an AI computing frenzy seemed to return. On last Thursday's U.S. stock market and Friday's Asia-Pacific stock market, the semiconductor sector, closely associated with AI computing infrastructure, staged a super rebound from "extreme leveraged position forced liquidations" to "revenge buying." When looking at the astonishingly strong performance and future outlook recently reported by leading players in the AI computing supply chain, such as Lam Research, Samsung Electronics, UMC, TSMC, SK Hynix, and Seagate, alongside South Korea's semiconductor exports soaring 179% year on year in July, the conclusion seems clearer: the physical demand associated with AI computing infrastructure has not deteriorated alongside stock prices and the extreme leveraged positions that plummeted and were cleared. The viewpoint is cutting-edge, but the positions collapsed first! The "AI oracle" faced Wall Street's oldest liquidity judgment. "This month we disappointed you," Leopold Aschenbrenner wrote in a letter to investors. However, even after experiencing a remarkable surge before the sell-off, the Situational Awareness funds investment returns were still up about 80% this year. The first to collapse was not the investment logic but the financing structure. This moment carries a somewhat unsettling echo of Wall Street history. The two tech booms followed strikingly similar paths in their respective initial four years both nearly erupted into a hedge fund leverage crisis at almost the same point in time. Bespoke Investment Group charted the performance of the Nasdaq index post the launch of Netscape in 1994 and after the explosive emergence of ChatGPT in 2022. The two tech booms displayed astonishingly similar trajectories in the early years, while the liquidation of positions by Situational Awareness coincidentally occurred around the same time as the 1998 crisis that nearly toppled Long-Term Capital Management (LTCM). The critical difference lies in the fact that Situational Awareness never constituted a systemic threat anywhere near the scale of Long-Term Capital Management. Their similarities were more about fundamental levels. Brilliant AI geniuses discovered a powerful concept. They concentrated bets on the highly focused theme of AI computing investments. Subsequently, leverage allowed timing to dominate. Aschenbrenner recognized the potential scale of the AI data center construction wave early on. In his forward-looking white paper titled "Situational Awareness: The Next Decade," published in 2024, he depicted the potential for exponential growth in global computing power, delving deeply into future projections along these trend lines. "Large American corporations are preparing to invest trillions of dollars, initiating a mobilization of American industrial power not seen in years," he wrote in this forward-looking report. As new data centers are built, power agreements signed, and overall capital expenditure plans for AI computing infrastructure hitting new highs, this prediction no longer seems as aggressive. The AI investment craze has begun shifting corporate cash flows toward AI chips, server clusters, high-performance network equipment, and data center power equipment. But measured over the course of a year, investment cycles have collided with an unprecedented leveraged financing structure that may be tested within a matter of days. A decrease in asset prices could trigger margin callsforcing investors to sell assets before their investment logic is validated. Aschenbrenner's background is as follows. Leverage allows investors to borrow funds, thereby controlling larger portfolios. When asset prices rise, borrowed funds amplify returns. This mechanism runs just as quickly in reverse. A decline in held asset prices decreases the value of collateral for the loans. Lenders may demand more cash or reduce the size of portfolios. This can force investors to quickly sell liquid assets, regardless of whether the initial investment logic has changed. Situational Awareness indicated that rapid declines in asset prices and weakened market liquidity made it increasingly difficult for the fund to maintain its portfolio within risk limits. Aschenbrenner likened this dynamic to a bank runevery sign of vulnerability creates more vulnerability. The hedge fund ultimately sold off most of its public equity investments and completely eliminated its leverage. It retained private market investments, which included its holdings in Anthropic. The assets sold from the portfolio weren't necessarily the least favored by Aschenbrenner, but rather those that were easiest to liquidate quickly. Public market stocks trade all day long and have constantly changing market prices. Private holdings may be harder to sell, but they also do not provide lenders with an instant path to cash. When pressure rises, liquidity may dictate which assets are retained and which must be sold. Wall Street has seen this movie before, although the assets playing the lead role have changed. The classic history of leveraged blowups on Wall Street is shown below. Concentrated positions decline. Borrowed funds amplify losses. Demands for cash collateral follow in succession, and investors ultimately lose control over the timing of their exits. Different markets, the same trap. This also explains why the brutal costs of leverage affect more than just hedge funds. Leveraged ETFs, margin accounts, or options positions can create the same misalignment between investor timelines and market timelines or timelines. Aschenbrenner's investment logic regarding AI may ultimately prove correct. Unwinding the leverage has bought the fund more time to validate this. But Wall Street will not reward you with more time just because you are smart enough. Long-term investment logic can wait; however, lenders will not wait for you. The profound conclusion of this myth's end is that Aschenbrenner may have accurately seen the long-term direction of the AI computing industry, but he mistakenly equated technological certainty with stock price certainty. HBM, NAND, advanced packaging, CPUs, optical interconnect devices, and power and liquid cooling demand may continue to grow, but an excellent investment strategy must be able to withstand valuation compression, shifts in correlation, tightening financing conditions, and crowded position reversals. The core of the Kelly Criterion is not that "high win rates should warrant heavy bets," but rather that it maximizes long-term compounding while avoiding bankruptcy; once the positions exceed the optimal proportion, even if the ultimate industrial judgment is correct, short-term variance can be sufficient to force an investor out before the theme can be realized. Situational Awareness has proven the most brutal rule of this round of the AI bull market: the market rewards not those who see the future first, but those who both see the future and have sufficient liquidity to survive until the future. "This is certainly part of the ongoing sell-off pressure that has existed for the past few weeks," said Calvin Yeoh, co-manager of Merlion Fund under Blue Edge Advisors, discussing Leopold Aschenbrenner's hedge fund's predicament. "At the same time, all retail investors using extreme leverage strategies in South Korea have essentially gone bankrupt; there are no larger sellers left." After the fall of the AI oracle, has the familiar feeling of the stock market frenzy around AI computing finally returned? Citadel took on most of Situational Awarenesss public equity positions, compounded by a record influx of foreign capital into South Korea, tightening of leveraged ETF regulations, and soaring semiconductor exports, which significantly suggests that the most dangerous "forced deleveragingmargin callindiscriminate sell-off" negative feedback loop of this round of AI computing chains has clearly eased; but it more accurately marks a strong recovery supported by fundamentals, rather than confirming a new round of indiscriminate semiconductor bull market frenzy. In other words, the end of deleveraging is beneficial for forming a market bottom, but it does not mean that a new round of leveraged prosperity has begun. To genuinely confirm the main rising wave of semiconductors, major semiconductor benchmark indices need to sustain above key technical levels without breaking during pullbacks and adjustments, alongside a continuous inflow of net funds and upward revision of profits in the coming weeks, while funds from leveraged ETFs in the South Korean and U.S. markets must stabilize without sharp outflows, and cloud vendors such as Microsoft and Amazon must continue to prove that capital expenditure can be transformed into cloud revenue, order backlogs, and acceptable asset recovery periods. Situational Awareness faced margin calls due to concentrated, leveraged bets on AI stocks, forcing it to transfer most of its public equity portfolio to Citadel; subsequently, Wall Street traders attributed part of the technology stock rebound on July 31 to the removal of this major selling pressure source. Its microstructural significance is very clear: the exit of a price-insensitive large seller halted the pressure from primary brokers to ensure collateral, and other investors no longer needed to sell ahead of them; this led to short covering and market makers' hedging, which in turn amplified the rebound. The strong trajectory of the fundamentals of the AI computing infrastructure industry is providing more persistent fuel for this semiconductor rebound than short covering. The most compelling fundamental evidence of this rebound comes from South Korea's July exports. Total exports from South Korea increased year on year by 62.8% to $98.89 billion, exceeding market expectations; specifically, semiconductor exports surged 179%, and computer exports jumped 404%, primarily driven by AI data center investments, rising storage chip prices, and SSD demand. Exports not only continued to grow strongly but also showed significant increases to the U.S. and China, indicating that global demand for AI hardware has not weakened in tandem with the stock price plummet in July. Microsoft's latest quarterly Azure revenue surged 43%, surpassing the market expectation of about 39.98%, showing that AI capital expenditure has begun to turn into cloud revenue and cash flow; Amazon AWS revenue grew 37% to $42.2 billion, significantly above market expectations, with order backlogs rising from $364 billion to $496 billion, and management stated that most of the computing capacity for 2027 has already been booked by customers; even raising capital expenditure for 2026 to $220 billion will not fully meet demand. Lam Research announced an average revenue guidance of $8.1 billion for the next quarter, clearly higher than the market expectation of $7.09 billion. These latest supply chain data are proving that AI demand is propagating toward HBM, DRAM, NAND, advanced packaging, etching/deposition, and enterprise-grade SSDs alongside GPUs. From the perspective of Wall Streets major firms, as this round of forced deleveraging nears its end, there are signs that a new super bull market themed around AI computing has "restart indications," but it cannot yet be declared that the main rising wave has been fully confirmed on the funding side. After Situational Awareness cleared the leverage, it demonstrated that the most severe mechanical selling pressure may have passed, but it cannot prove that all over-leveraged institutions have been cleared; even with the KOSPI surging nearly 18% in a single day, it is still about 30% lower than its historical peak, and the Philadelphia Semiconductor Index remains down over 20% for July, having previously retraced nearly 30% from June highs. The familiar atmosphere of expanding AI risk appetite, with a vibrant and competitive bullish sentiment toward AI computing fundamentals, may have returned, but the trajectory of the primary rising wave of an AI-driven super bull market still requires continuous confirmation through financial reports, orders, returns on AI capital expenditure, and holding technical lows during pullbacks. JPMorgan stated that the long-term bullish logic supporting South Korean stocks and semiconductor stocks related to the AI computing theme is the near-endless, ongoing expansion of AI data centers for HBM, server DRAM, enterprise-grade SSDs, and advanced storage, compounded by discipline in supply which still supports spot and contract prices. With foreign capital returning to the South Korean stock market last week, setting a historical single-day record, Morgan Stanley's publicly released "Buying the AI Infrastructure Dip" report attributed the recent declines in AI infrastructure stocks mainly to short-term positioning, crowded trades, and technical factors, rather than a deterioration in demand fundamentals. Its core judgment is that demand for AI computing may continue to far exceed supply over the next few years; more efficient, cost-effective models may paradoxically broaden token calls and total computing demand per Jevons Paradox; constraints around electricity, skilled workers, and approvals serve as "speed bumps" that slow construction, rather than "brick walls" that halt industry expansion. Morgan Stanley predicts that the overall capital expenditure of the five major tech giants in the U.S. will rise from nearly $800 billion in 2026 to about $1.2 trillion in 2027 and approximately $1.4 trillion in 2028. This provides exceptionally high revenue visibility for assets related to AI computing infrastructure, such as GPUs, HBM, Ethernet switch chips, optical communication/optical interconnects, cooling infrastructure, data center CPUs, and power equipment, among others. At the same time, such enormous capital expenditure also means that the market's next stage will not only reward "getting orders and outlooks," but will rigorously scrutinize AI revenue increments, free cash flow, unit token costs, asset depreciation lifespans, and financing costs. In Morgan Stanleys view, the long-term bull market for AI computing infrastructure has not ended, but the best opportunities after a correction are concentrated in companies that genuinely control bottlenecks in computing power, storage, electricity, and data center delivery, rather than all high-beta assets tagged with AI.