Before getting to the leverage, it is worth introducing Leopold Aschenbrenner.
For those who do not know him, Aschenbrenner is a former member of OpenAI’s Superalignment team who previously researched economic growth at Oxford’s Global Priorities Institute. He later founded an hedge firm focused on AGI. In 2024, he published Situational Awareness: The Decade Ahead, a series of essays on how rapidly advancing AI could reshape technology, industry, energy, national security and geopolitics.
I recommend reading it.
You do not have to agree with his AGI timeline, his geopolitical framing or every conclusion. I do not. But it is one of the rare AI publications that follows the technology all the way down to the physical systems required to support it.
Aschenbrenner did not write about AI as if it were weightless software floating somewhere in the cloud. He wrote about GPUs, data centres, power plants, chip fabs and the enormous amount of capital that would have to be mobilised.
In Racing to the Trillion-Dollar Cluster, he argued that many trillions of dollars would flow into GPU, data-centre and power infrastructure before the end of the decade. He also identified power, land, permitting and construction as likely constraints, long before these topics became part of almost every AI-infrastructure conversation. Whether his exact timeline proves right or not, that section now reads less like science fiction and more like an early map of the infrastructure race unfolding around us.
Leopold Aschenbrenner saw the infrastructure consequences of AI earlier and more clearly than almost anyone.
Then he put money behind the thesis. In layman's terms: He put his money where his mouth is.
His fund, Situational Awareness, concentrated heavily on the companies expected to provide the chips, memory, energy and infrastructure behind the AI buildout. It returned 439% in the first half of 2026 and grew to more than $20 billion. For a while, Aschenbrenner looked like the person who understood the moment better than the market itself.
Then July reversed it.
Several of the fund’s concentrated positions fell sharply. The public portfolio was reportedly running at roughly four times gross exposure, magnifying the pressure as the market moved against it. Margin calls followed, and the fund sold most of its public-equity portfolio to Citadel. It lost 67% in one month and was left with an estimated $10 billion, much of it in private holdings. Remarkably, it was still approximately 80% up for the year.
Here is the part worth sitting with:
None of that, by itself, disproved his thesis.
The demand he anticipated is still real. The infrastructure buildout is still happening. Many of the companies he invested in remain central to where AI is going. What failed was not necessarily the destination. What failed was a portfolio structure that could not survive the journey. And the two are not the same thing.
There is a line widely attributed to Charlie Munger that there are only three ways a smart person goes broke: liquor, love and leverage. Warren Buffett later expressed the part that really matters:
“History tells us that leverage all too often produces zeroes, even when it is employed by very smart people.”
That is the whole story here.
Liquor and love can take down all kinds of people. Leverage has a particular talent for taking down smart ones, because you generally need to be credible enough to be trusted with it in the first place. It is the failure mode reserved for people who are right often enough that they begin to structure their positions as if being right is the only variable that matters.
Aschenbrenner may still be right about the destination.
Leverage did not wait for the verdict.
There is a real lesson in this, and it is not that Aschenbrenner was a fool. He was not.
The lesson is that being right about the future does not protect you from the way you structure the bet.
A temporary drawdown can be survivable. Add enough leverage and the same drawdown can become permanent. The thesis may not have changed at all, but you are forced out before you get the chance to find out whether you were right.
And this is where I think two things can be true at once.
First, leverage is dangerous. It can turn a temporary market move into permanent capital impairment, even for very smart investors.
Second, Aschenbrenner’s underlying AI infrastructure thesis may still be right. The GPUs still need to be built. The power still needs to come from somewhere. Data centres, chips, networking and energy remain fundamental constraints on how quickly AI can scale.
His fund blowing up does not prove the thesis wrong any more than the 1,000%+ rise beforehand proved it right. Those are two different questions. The performance of the investment vehicle tells us something about leverage and risk management. The next several years will tell us whether Aschenbrenner was right about AI infrastructure. I would not bet lightly against him on the second one.
So the takeaway is not that AI infrastructure was a bubble, nor that Aschenbrenner was a fool.
It is simpler:
Two things can be true at once. Leverage can destroy a great position, and the thesis behind that position can still ultimately prove correct.
And he was reminded, very expensively, of Munger’s third L.