AINEWS
Concentrated AI funds lost 6.8% to 21.7% in July on the same chip stock bet, while AI software rallied, a hardware only repricing inside the broader AI complex.
On July 1, Whale Rock Capital was up 72.5% for the year. By August 1, that number had been cut to 35.1%. The Boston-based concentrated value fund, run by Alex Sacerdote, lost 21.7% in July alone, according to triangulated reporting from QbitAI, Business Insider, Seeking Alpha, and Briefs.co. A single month erased half of the fund's 2026 gains.
Point72's Turion fund, a long/short vehicle focused on AI hardware and chip companies, fell 11.4% in July. Coatue, one of the most-watched tech-focused hedge funds, dropped 8.3% in its worst month in over a year, cutting its year-to-date return to 14.3%. Eureka, another AI-focused shop, slid 6.9% to 11.6% YTD. The pattern was clean enough to name: in July, the funds that bet most heavily on AI got hit together.
What they were all holding was the same trade. SK Hynix, the Korean memory-chip maker whose high-bandwidth memory (HBM) is the bottleneck input for every advanced AI training cluster, fell roughly 40% from its June peak. SanDisk, Micron, and CoreWeave, a neocloud operator that rents GPU capacity to AI labs, all dropped more than 30% in July. The lever these funds were pulling was the AI infrastructure complex, not AI software. Adobe, Salesforce, and the application-layer names actually rallied in the same window, a split that defines what happened in July. Turion was one of the defining AI-infrastructure trades of the boom, and the chip names it rode cratered together.
The market wasn't rejecting AI. It was repricing the AI infrastructure-payback story specifically. Two things converged. The scale of the bet is staggering: Google, Amazon, Meta, and Microsoft have collectively announced roughly $720 billion of 2026 capital expenditure, almost all of it earmarked for data centers and the GPUs, networking, and memory that fill them. Separately, a pair of Chinese open-source model releases, K3 and V4 Flash, reignited the question of whether US infrastructure is being over-built relative to the AI demand that actually monetizes. K3 and V4 Flash are open-weight models, meaning anyone can download and run them, which compresses the moat that the US labs' closed frontier models have been selling to investors. Chinese-language coverage, including a 163.com analysis, framed it as the "compute gap is over" story.
Leopold Aschenbrenner's "Situational Awareness" fund became a sidebar to the same story. The 24-year-old former OpenAI researcher's fund, named for his widely-read 2023 essay series on US-China AI competition, reached $1.5 billion in assets in roughly a year and was reported up 439% in the first half of 2026, before collapsing in July under what an LP letter described as high leverage on the same hardware names. The fund's trajectory compressed the broader pattern into a single portfolio: it worked, until it didn't, and then it worked in reverse faster than almost anything else on the public-fund scoreboard.
The diagnostic worth keeping: in July, software held up, hardware didn't, and the spread between the two legs of AI widened into a tradeable gap. If the next capex print from Google, Amazon, Meta, or Microsoft shows discipline, or if the next open-source release from a Chinese lab lands without a comparable US response, the hardware leg has a reason to recover. If software starts selling off while hardware stabilizes, the story has changed. The July split is the working model until something breaks it.