Source Foundry Gets $400M From a Fund Down 67%

Two weeks after the worst month of its short life, the AI hedge fund Situational Awareness has put another $400 million into a chip startup almost nobody has heard of. The investment in Source Foundry, first reported by The Wall Street Journal and relayed by TechCrunch, takes the fund's total commitment to $500 million and values the company at roughly $5 billion.

Source Foundry launched last year out of Stanford, co-founded by researchers Abdulmalik Obaid and Joe Burg, and it is aiming at the narrowest point in the entire AI supply chain: extreme ultraviolet lithography. Every leading-edge AI accelerator — Nvidia's, Google's, Amazon's, Anthropic's forthcoming custom silicon — is patterned by an EUV machine, and every EUV machine in the world is built by one company, ASML of Veldhoven. The tools cost north of $400 million each, arrive in pieces the size of a school bus, and take months to install. ASML booked €32.67 billion in sales in 2025 with no direct competitor.

That is the bottleneck Source Foundry, which also counts Sequoia Capital among its backers, says it can widen — building machines, equipment and software to make advanced chip manufacturing faster and cheaper. It is an unusually deep place to attack. Everything downstream has been scaling on demand: TSMC posted a record $39.6 billion quarter on AI orders, and model builders are now commissioning their own accelerators, as with Anthropic's reported move to Samsung's 2nm line. None of that expands unless the tools that print the wafers expand first.

The investor is the strange part of this story. Situational Awareness was founded in 2024 by Leopold Aschenbrenner, a former OpenAI researcher then in his early twenties with no prior trading experience, on a straightforward thesis: buy the companies that benefit if AI capability keeps compounding. It worked spectacularly, then stopped. The fund fell about 67% in July as AI infrastructure stocks sold off, a drawdown severe enough that Mike Novogratz called it the most catastrophic hedge fund collapse he had seen in his career. Late in the month it sold most of its public book to Ken Griffin's Citadel to cut leverage, keeping its Anthropic stake, and its assets under management went from roughly $20 billion to $10 billion.

So the same fund that just discovered how violently a public AI portfolio can reprice is concentrating what remains into an illiquid, pre-revenue private bet several layers upstream. There is a coherent argument for it. If your thesis is that compute demand keeps climbing, the safest expression of that thesis is not the chip designer whose multiple can halve in a quarter — it is the physical chokepoint that has to be relieved either way. Sequoia's Stephanie Zhan has made the same point about the upstream layers of the stack tightening as everything below them scales.

The counterargument is that lithography is where hardware ambitions go to die. ASML's monopoly is not a distribution advantage or a software moat; it is three decades of optics, plasma physics and a supply chain of sole-source components, much of it inherited from Zeiss. Intel, Canon and Nikon all have credible reasons to want an alternative and none has produced one. A $5 billion valuation on a company founded in 2025 prices in a lot of physics that has not happened yet.

What makes this worth watching is not the size of the cheque but where it landed. Capital chasing AI has spent three years moving steadily upstream — from apps to models, models to chips, chips to power and now to the machines that make the chips. Source Foundry is the end of that road, and $400 million from a badly bruised fund is a reasonable marker of how tight the road has become.