After a 67% loss, Situational Awareness backs chipmaking bottleneck

Leopold Aschenbrenner is putting another $400 million behind artificial intelligence just days after his hedge fund suffered one of its steepest setbacks.

The Wall Street Journal reported on August 7 that Situational Awareness has made an investment of $500 million in Source Foundry, which includes the company investing $400 million this week itself. Source Foundry started its operation in 2025 with Stanford researchers Abdulmalik Obaid and Joe Burg, and it has recently been assigned a market valuation of about $5 billion owing to the funding it has received from Sequoia Capital.

The timing makes the deal more significant than a conventional venture investment.

According to a letter to investors, Situational Awareness suffered an unaudited net month-to-date loss of 67% in July. While its net year-to-date returns were still at 80%, Aschenbrenner revealed that they sold part of their public portfolio to cut down on leverage following low liquidity conditions in the market.

However, the letter also clarified that the July fiasco did not alter his long-term view of AI. The fund will keep working as a hybrid public-private vehicle, while its public portfolio will be managed on a fully paid-for basis from now on, thus eliminating all margin or liquidation risks.

In light of this situation, the Source Foundry investment appears less as a move back to the strategy behind the losses of July than as a long-term commitment to the physical infrastructure required for AI development.

Aschenbrenner is betting on the bottleneck behind the chips

Source Foundry is far from just another designer of AI chips. The company aspires to be much more, focusing on developing machines, equipment, and software for semiconductor manufacturing, with an emphasis on lithography. Sequoia partner Stephanie Zhan has characterized its mission as having to do with the “tightest bottleneck” of semiconductor manufacturing, starting with lithography.

That means the startup will be competing with ASML, whose extreme ultraviolet (EUV) lithography technology is vital to manufacturing the world’s most sophisticated chips.

The technical challenge for Source Foundry is massive. ASML’s EUV systems use 13.5-nanometer light, which is produced when lasers are fired at small drops of molten tin. The light travels through a vacuum and is controlled by extremely accurate multilayer mirrors since regular lenses are unable to transmit EUV light in an effective manner.

There are no indications yet of how Source Foundry’s technology is different from ASML’s EUV system. But it is certain that the start-up is working on an alternative to today’s large, costly and complicated equipment for producing semiconductors.

In order for this endeavor to be successful, the technology being developed should provide more than just a machine capable of operating in a laboratory. It would need to be able to perform at a level that would satisfy precision, throughput, and reliability sufficient for mass production of chips while making progress regarding production costs, size or deployment time.

Thus, Source Foundry’s success will have implications beyond that of just one AI-chip manufacturer since every advanced processor maker relies on semiconductor manufacturing equipment.

The demand keeps rising. TrendForce predicts that global foundry revenue will increase by 24.8%, climbing up to $218.8 billion in 2026, with the demand for AI chips keeping advanced node capacity levels constantly high.

Put another way, what happened in July did not take away the actual need for a physical infrastructure for AI. It revealed the disadvantages of financing this demand via leveraged positions in the public market.

The bigger signal for the global AI market

The significance may go far beyond situational awareness.

As AI infrastructure grows, bottlenecks can shift further upstream—from models and accelerators down to memory, energy, advanced packaging, and eventually the machinery required to make the chips themselves.

An analysis conducted by both the Semiconductor Industry Association and Deloitte has found that the value of semiconductors in an AI server rack is more than 95%. Moreover, AI data center infrastructure investment could be more than $4 trillion by 2028.

This presents a note of contradiction for investors: the stocks of companies involved in AI can dip despite the existence of a solid infrastructure cycle.

Source Foundry is basically a gamble on what transpires when demand meets a much larger limitation: the machinery necessary to produce enough advanced chips.

It also underscores a lesson from Aschenbrenner’s July losses. Being right about AI’s long-term growth is not necessarily enough to survive short-term market volatility.

His 67% drawdown in July showed how leverage and concentrated positions can quickly turn harmful when markets change drastically, and liquidity disappears. Meanwhile, the $400 million investment in Source Foundry indicates that he has not compromised his thesis regarding artificial intelligence. He is merely focusing more on the less-known—and possibly key—part of the technology stack.

The question now is whether Source Foundry can transform that insight into actual semiconductor-processing technology. If it can do so, Aschenbrenner’s latest wager may be of even greater significance for AI than his lost public-markets venture, given that a breakthrough in chip-making devices would be beneficial and profitable not just to a single maker of AI accelerators, but to the whole industry.