AI investing risk: being right isn't enough

Leopold Aschenbrenner was about as close to artificial intelligence as an investor could get. Within days, most of his hedge fund's stock portfolio had been sold in an overnight deal. The AI investing risk the episode exposes has nothing to do with whether he was right about the technology.

Late on Wednesday night, Leopold Aschenbrenner got on the phone to Ken Griffin. By dawn, Griffin's Citadel had agreed to buy the bulk of the public stock holdings of Situational Awareness, the hedge fund Aschenbrenner founded in 2024 and built to $24bn in assets. Those holdings were once estimated at $16bn, and the Financial Times called the deal one of the largest and most sudden stock transactions in Wall Street history. The firm survives and keeps its private stakes, but the borrowed, concentrated equity strategy that produced its returns is finished. Aschenbrenner is 24 years old.

The obvious question, and the one the FT's reporting ends on, is how anyone came to invest and lend so much money to someone with no trading experience. It is a fair question. It is not the one that matters most to a reader with a pension or an ISA, because what ended that strategy was not that Aschenbrenner had been wrong about AI.

He was as close to AI as an investor could get

Aschenbrenner graduated from Columbia at 19 and joined OpenAI's Superalignment team, before being fired over an alleged leak. His fund held a $5bn stake in Anthropic. In 2024 he published a 165-page essay, Situational Awareness, arguing that AI would reshape society within a few years. It gave him a following in Silicon Valley and on Wall Street and more than 250,000 followers on X.

If an informational advantage on the direction of AI existed anywhere, it existed here. And nothing that has happened in the past fortnight tells us his thesis was mistaken. AI may yet transform the economy on roughly the timetable he set out.

But knowing where a technology is going is not the same as knowing what the market has already paid for it. A sell-off tells you a good deal about valuations and investor behaviour and very little about whether the underlying technology works.

What actually forced the sale

Situational Awareness borrowed heavily to increase its returns. Goldman Sachs and JPMorgan Chase were among its prime brokers. At least one prime broker had placed the fund on a watch list because of the volatile and concentrated nature of its AI positions. According to an executive at the bank, it had called for additional margin several times over the past year, including this week.

When Intel reported strong earnings on 23 July, its share price fell, and the FT reports that the aggressive seller in the market was Aschenbrenner, trying to recover his losses. One person familiar with the matter said he 'had the illusion of still being in control'.

A letter went to investors the next day setting out returns of 439 per cent for the year to the end of June and 1,551 per cent since the fund launched. Those were his own figures, not audited ones, and the letter invited investors to add money on 1 August. Within days he was ringing round soliciting capital and offering different terms to different people. By Wednesday he had stopped taking their calls.

Leverage is usually described as a way of magnifying returns. What it also does is hand somebody else the decision about when you sell.

The AI investing risk that isn't about AI

Almost no private investor has a prime broker, and the AI investing risk this episode illustrates does not depend on having one. Anyone can be made a forced seller: by redundancy, by a house purchase, by needing to draw an income in retirement. Borrowing simply makes the deadline explicit and brings it forward.

What turns a deadline into a loss is concentration. A portfolio spread across thousands of companies can absorb a bad month in one theme. A portfolio built on a single conviction cannot, and the evidence on concentrated funds suggests conviction more reliably produces downside than return. Aschenbrenner's positions in Bloom Energy and Sandisk had each fallen about 40 per cent between the end of June and this week.

What the story does and doesn't settle

It is not over. One person close to the firm said as much this week.

Hubris is somewhere in this story, but it is not the part that transfers. What transfers is that a correct view of a technology and a profitable position in it are separate things, which is why transformative technologies have so often rewarded their users rather than their investors.

Aschenbrenner may still be proved right about AI. He will not be paid for it.

Resources

Aliaj, O., Shah, J. R., Hammond, G., Pollard, A., Mourselas, C., Duguid, K., & Fontanella-Khan, J. (2026). How Leopold Aschenbrenner, the 'golden child' of the AI trade, was laid low. Financial Times.

Aliaj, O., Pollard, A., Mourselas, C., Duguid, K., & Fontanella-Khan, J. (2026). Citadel buys Situational Awareness equity holdings after steep AI losses. Financial Times.

If you're not sure how concentrated you are

Most people discover how much of their portfolio sits in one theme only when that theme has a bad month. An evidence-based adviser can help you identify that exposure beforehand, and match your portfolio to when you are actually likely to need the money. TEBI's Find an Adviser directory lists firms we're happy to recommend.

The evidence, by email

The TEBI Digest is our free email newsletter on investing, financial planning and behavioural finance. To sign up, head to evidenceinvestor.com and enter your email address in the 'Subscribe' box at the foot of the page.