Listen to this article
Leopold Aschenbrenner built Situational Awareness into the hottest hedge fund of the artificial-intelligence boom following his viral essay “Situational Awareness: The Decade Ahead”. Then came the past month: his fund’s portfolio fell 67 percent as its concentrated, leveraged bets on AI stocks reversed, forcing it to sell most of its public-equity holdings and eliminate its leverage. The episode has been treated mainly as another standard tale about the pitfalls of excessive leverage (too much borrowing, which magnifies returns). It may also be a warning about the tech-driven market around him. Now the SEC is investigating the firm.
Leverage certainly explains the violence of the fund’s decline, as a fund operating with four or five times leverage can transform an ordinary correction into a catastrophe. But leverage does not explain why the underlying AI stocks became so volatile in the first place. Situational Awareness’s collapse may be the first major crack in a market that has priced an extraordinary (an uncertain) technological future while assuming unusually favorable financial conditions.
Aschenbrenner may still be right about the general trajectory of AI technology. I think we’re all right to be very optimistic about it, without being “irrationally exuberant,” as Alan Greenspan described the tech stock overconfidence of the dot-com era. Generative AI is transforming software, scientific research, and business operations, raising productivity and economic growth. The prospects for the American economy look terrific as they did in the 1990s during the dot-com revolution. Yet investors are confusing a powerful technological thesis with the proposition that every AI-related security is appropriately priced. A great technology can make a bad short-term investment when optimism, leverage, and low discount rates are already priced in.
Tech stocks are valued highly and could soon face a short-term reckoning once the Fed begins hiking rates much more this Autumn. Think about the last rate hikes in 2022, which caused a significant selloff in tech stocks (on average they sold off more than 25%). This isn’t to say that AI isn’t a technological revolution that will meaningfully help drive productivity and economic growth in the future, just that there will be some bumps along the way.
Fed-funds futures currently put the odds of a September rate increase by December at roughly fifty-fifty, or seventy percent. In my view, that is an underestimate. Chairman Kevin Warsh has repeatedly said that the central bank has no tolerance for inflation above two percent, yet PCE inflation has remained above the Fed’s two percent target for over five years. Some market commentators feel such comments at the past few meetings lack credibility since the Fed did not go through with any rate hikes, nor signal much about the future, a critical piece of Warsh’s stance.
As the federal funds rate (the key benchmark interest rate target for the Federal Reserve, which is the rate at which banks lend to each other) is currently between 3.50 percent and 3.75 percent, this is close to the median estimate of the neutral rate of interest that market participants find. This means that if the Fed wants to meaningfully bring down inflation, it must raise interest rates from current levels.
That complacency arguably matters most for technology stocks. Their valuations depend heavily on earnings expected far into the future, making them unusually sensitive to the interest rate used to discount those earnings. The 2022 selloff was not caused by a collapse in the usefulness of software or cloud computing. It reflected the rapid repricing of long-duration assets as the Federal Reserve prepared to raise rates, a repricing the market was slow to price in.
The fiber-optic boom offers another useful comparison. Investors in the late 1990s correctly predicted that the internet would require enormous networks of fiber-optic cable. Companies raced to lay cable across the country and beneath the oceans, building infrastructure that ultimately became indispensable. Yet Global Crossing went bankrupt, and Level 3 Communications inflicted large losses on shareholders, because investors had paid prices that assumed demand, financing, and profits would all arrive on the most favorable schedule.
The fiber was valuable (we still use much of the same fiber today), even if many of the stocks were not priced at what investors paid. Many of today’s tech companies, including the AI companies, are built on that very same fiber.
AI could very well follow the same pattern. The technology may exceed today’s grandest forecasts while its stocks suffer a sharp interim correction. Strong AI investment could itself keep demand, electricity consumption, and capital spending elevated, giving the Fed more reason to remain restrictive.
Situational Awareness’s fall from grace should therefore be understood as more than a leverage accident. It illustrates how quickly crowded AI trades can unravel when financial conditions turn less forgiving. Aschenbrenner’s mistake may not have been believing too much in AI. It was financing that belief as though the cost of money no longer mattered.
The broader market may be making a similar error. Investors may have situational awareness of the technological future, but they remain unaware of monetary policy and the Federal Reserve.
Jon Hartley is a Research Fellow at the Civitas Institute, and an Assistant Professor of Economics at the University of Texas at Austin School of Civic Leadership
