Unlock the Editor’s Digest for free
Roula Khalaf, Editor of the FT, selects her favourite stories in this weekly newsletter.
Leopold Aschenbrenner was hailed as a great prognosticator, but he did a poor job of interpreting history. The tech wunderkind’s misfortune is a reminder — useful for the AI era — that an investor can be right and still come undone.
Situational Awareness, the fund the OpenAI alumnus founded in 2024, sold most of its publicly traded securities to rival Citadel after banks called in their loans. On the face of it, what happened is simple: Aschenbrenner thought AI-related stocks would go one way; they went the other. Since he’d funded his trades with debt, this quickly turned into a liquidity crunch.
Was he wrong? To mix short-term leverage with volatility, definitely. It ought not to be a shock if what went up like a rocket sometimes comes down like a stick. Of the 29 US-listed stocks the fund reported holding at the end of March, the average share price is down 21 per cent in the last month. Chipmaker Sandisk, one of its large holdings, is down 45 per cent.
That doesn’t mean Aschenbrenner wasn’t directionally right. He claims the fund is still 80 per cent up this year. Situational Awareness still holds unleveraged stocks, and a stake in yet-to-go-public AI lab Anthropic. It can now weather valuation wobbles without having to dump those crown jewels, having learnt the hard way that debt and unpredictability don’t mix.
Who else still needs to learn that lesson? Perhaps builders of AI data centres — who Lex reckons are on course to invest $9tn by 2030, and are at risk of making similar mistakes, albeit played out in years rather than days. Just as Aschenbrenner’s long-term leveraged bet proved short-term wrong, builders of tech infrastructure, taking on boatloads of debt financing, are shoring up what Man Group warns is a “temporal mismatch.”
Plenty of things can go wrong. Chips could wear out before the loans secured against them have been repaid. New AI model efficiencies could make some data centres superfluous. Chinese models such as Kimi K3 that do more with less could gain traction. In sum, the view that AI will gain mass adoption could be totally right, but the returns, specific winners and timeframes could vary wildly, bringing calamity for some with lenders to appease.
In his defence, Aschenbrenner was in his infancy when the financial crisis hit. His banks, though, were not — so it’s not surprising they pulled back when the market turned against him. Nonetheless, while big lenders have avoided betting their own balance sheets on the AI boom, they remain happy to help others do so, for a fee. The lessons of history are there for those who heed them, and leverage is still available for those who don’t.
[email protected]
