Risk gaugegettyA sharp July unwind in AI‑linked hedge‑fund positions revealed a risk that has little to do with artificial intelligence itself: when too much capital crowds into the same trade and finances it with borrowed money, the investment horizon can collapse overnight. The selloff showed how quickly leverage can turn a long‑duration thesis into a forced, short‑term decision.Leopold Aschenbrenner built one of the most widely discussed investment records of the AI stocks in the boom by seeing the scale of the computing buildout earlier than most investors. His hedge fund, Situational Awareness, reportedly returned 439% during the first half of 2026 and grew to around $20 billion. Then a sharp July selloff in AI-related companies contributed to losses, demands for additional capital and conversations with investors and lenders. The fund later sold a large part of its public equity portfolio to Citadel. There will be plenty of debate about whether Aschenbrenner became too confident, whether the AI trade had become overcrowded or whether the selloff exposed unrealistic valuations. I see a more useful investment lesson. Artificial intelligence’s long-term benefits were not necessary for the portfolio's success. The financing only needed to become uncomfortable.AI development will take years. Leverage is assessed daily. A lender does not care that a semiconductor cycle, data center expansion or private AI investment may produce enormous value by 2030. Collateral is measured at today’s price. When those two time horizons meet, the financing can end the investment before the thesis has had time to work.Why AI Stocks Can Turn Into One Crowded TradeSituational Awareness reported 42 entries worth approximately $13.68 billion in its March 31 Form 13F filing. The filing covered positions across semiconductors, data centers, memory, power and other infrastructure linked to AI spending. A 13F is only a quarter-end snapshot, and option positions can make the economic exposure harder to interpret, but the filing still shows how broadly the AI theme had been expressed.MORE FOR YOUThe filing showed exposure across major AI‑linked infrastructure companies, including Nvidia, AMD, Oracle, Micron, Broadcom, Taiwan Semiconductor, Bloom Energy and CoreWeave. Those businesses do different things. Their customers, margins and capital requirements are not the same. The market can still treat them as one position.If the common assumption is that AI capital spending will continue accelerating, then several apparently separate holdings may depend on the same source of confidence. Investors are not diversified merely because one company manufactures chips, another supplies memory and another provides computing capacity.The same money can own all three companies. We saw versions of this scenario during earlier investment cycles. In the technology bubble, investors owned networking equipment, fiber, software and internet companies and believed they had diversified across the digital economy. Much of the portfolio still relied on one assumption: that capital spending and valuations would continue rising together.The AI buildout may prove far more economically important than the internet businesses of that period. The portfolio risk is familiar. Different securities can become highly correlated once the same owners begin reducing exposure.This matters because a crowded position does not always unwind according to fundamental quality. Investors sell what is liquid, what has appreciated most and what can raise cash quickly. The best company in the portfolio can therefore fall alongside the weakest.AI Stocks And The Cost Of Borrowed ConvictionInvestors often describe leverage as a way to express greater conviction. That description works while prices rise. Once they fall, collateral becomes more important than conviction. A portfolio manager using borrowed money does not have complete control over the investment horizon. Prime brokers, lenders, redemption requests and internal risk limits begin to influence what can remain in the portfolio.The fund usually sells the most liquid holdings first because they can produce cash without waiting for a buyer. This can leave a fund holding more of its private and difficult-to-sell assets precisely when liquidity matters most.Situational Awareness retained private investments, including an early stake in Anthropic, after selling much of its public equity book. Anthropic listed Situational Awareness among the investors in its May funding round. That investment could eventually prove extremely valuable, but a private stake cannot be converted into cash as easily as a large publicly traded semiconductor position.That creates an uncomfortable result. The fund may keep the asset with the greatest long-term potential while losing the liquid positions that allowed it to finance and manage the portfolio around it. Long-Term Capital Management employed experienced investors and sophisticated models, yet leverage left the fund exposed when market relationships moved against it in 1998. The concern became an uncontrolled fire sale rather than whether every position would eventually recover.Archegos provided a more recent example. It used swaps and margin to build enormous exposure to a small group of companies. When several positions declined, hundreds of millions of dollars of excess margin disappeared within days, followed by demands the fund could no longer meet. The market then stopped caring about the original investment case. The stock had to be sold.Aschenbrenner’s situation is not identical to either case. The comparison is about structure. Concentration and borrowed money can turn an ordinary drawdown into a decision made by someone outside the investment team.Great Returns Can Hide Fragile FinancingA 439% six-month gain gives a manager credibility. It also changes the portfolio.Assets rise, investors add capital and lenders become more comfortable extending financing. Exposure can expand at the moment when recent performance makes the strategy appear safest. That is how risk hides inside success.An investment that rises fivefold has created wealth, but it may also have encouraged a larger position, more borrowing and a greater dependence on liquid markets. The previous gain does not protect the portfolio if the current exposure has grown faster than its ability to withstand a reversal. Investors tend to judge risk using the return from the starting date to the ending date. That misses what happened between them.A fund can report an exceptional long-term return and still become vulnerable to a relatively ordinary monthly decline. The final result depends on whether the investor remained in control long enough to capture the recovery.Downside protection is sometimes presented as caution. I see it as mathematics. A position that must be sold before the thesis develops never earns the return shown in the model. This concept becomes particularly important in AI because the capital cycle is extensive. Semiconductor factories, power generation, data centers and model development require investment today for demand that may arrive years later.Long-duration assets need patient financing. Borrowed money can remove that patience at exactly the wrong moment.What Forced Selling Means For AI StocksThe sale of a large AI portfolio does not tell us that AI capital spending has peaked. It does suggest that some of the ownership behind AI stocks was less permanent than it looked. That can create opportunity.When a forced seller is reducing exposure, price becomes secondary. The immediate aim is to raise cash and lower risk. A fundamentally strong company can fall because one of its holders no longer has the balance-sheet capacity to wait.Investors with cash and little leverage can benefit from that pressure. Some of the best opportunities in special situations appear when the seller is responding to financing, index changes, redemptions, or a mandate rather than a deterioration in the business. The work is identifying the reason for the sale.A falling stock does not provide that answer. Oracle and AMD reportedly fell about 20% during July, but investors still need to determine how much of the move reflected forced selling and how much represented a reassessment of earnings, capital expenditure, and valuation.Not every decline in an AI stock is a liquidity event. Some companies are priced for years of uninterrupted spending, expanding margins, and limited competition. If those expectations change, the market may justify the lower price. Ownership analysis therefore matters alongside company analysis. Who holds the stock? How much leverage is behind the position? Are investors committed to the business, or are they committed to the theme while prices rise?Aschenbrenner may still be right about the scale of the AI buildout. His private investments may eventually produce substantial value, and the July losses could look temporary several years from now. The portfolio has still delivered a warning.Artificial intelligence and AI stocks require investors who can survive long periods between capital spending and economic return. Once leverage shortens that period, the lender can become more important than the thesis.