Berkshire's $339 Billion Treasury Position Is the Bear Case on AI Capex That Buffett Won't Say Directly

The framing circulating this weekend, that Warren Buffett has judged AI to be overvalued, oversimplifies the record, but the underlying numbers are real and worth an engineer's attention because they price the opportunity cost of the current build-out. Berkshire Hathaway ended the first quarter of 2026 with $397.4 billion in cash and short-term instruments, of which roughly $339.3 billion sat in short-term US Treasury bills earning, at prevailing yields, on the order of $20 billion a year.

That is the comparison. While the largest AI companies commit hundreds of billions of dollars to models and data centers on the expectation of future returns, a risk-free Treasury position currently pays about 4 to 5 percent with no execution risk. Buffett has repeatedly said that cash is "not a good asset" over the long run but that he cannot find large acquisitions at sensible prices, which is a statement about valuations rather than a verdict on AI specifically.

The picture is more mixed than the headline suggests. Berkshire built a stake in Alphabet worth roughly $31 billion, including a $10 billion private purchase, which is a direct bet on an AI-exposed platform. The firm is therefore not avoiding AI so much as choosing where in the technology stack to take its exposure, favoring a cash-generative incumbent over the capital-intensive model layer.

Who benefits if the skeptical reading is right: holders of the hyperscalers and chip suppliers lose if AI revenue does not grow into the depreciation schedules that the capex implies, while a Treasury-heavy balance sheet benefits simply by waiting. Who benefits if it is wrong: the labs and their compute suppliers, if the spending compounds into durable demand.