The AI bubble argument goes mainstream

Through November and December 2025, the argument that AI infrastructure spending had become a financial bubble moved from analyst notes and short-seller commentary into mainstream financial press, anchored by investor Michael Burry’s public bet against AI hyperscalers and a wave of year-end reporting on the scale of the buildout.

Burry, known for his short position ahead of the 2008 housing crash, argued that Meta, Oracle, Amazon, Microsoft and Google were understating depreciation on Nvidia GPUs by assigning them accounting lifespans of five to six years when their practical useful life — given Nvidia’s roughly annual product cycle — was closer to two or three. He calculated this would understate depreciation, and so overstate earnings, by roughly $176 billion cumulatively between 2026 and 2028, with Oracle’s profits overstated by around 27% and Meta’s by around 21%. Nvidia disputed the framing, saying its chips remained productive well beyond critics’ estimates; CNBC reported it could not independently confirm the specific accounting claims.

A related strand of commentary focused on circular financing: Nvidia investing in OpenAI, which spends heavily on compute from Oracle and Microsoft, which in turn buy chips from Nvidia. Analysts, including Bernstein’s Stacy Rasgon, compared the structure to vendor financing in the dot-com era, when Lucent and Nortel lent customers money to buy their own equipment. By 19 December, CNBC reported data-centre dealmaking had reached a record $61 billion for the year, with hyperscalers increasingly turning to private-credit debt rather than balance-sheet cash to fund construction, even as “AI valuation risks and funding concerns grip investors.”

Proponents of continued investment countered that demand for compute remained undersupplied and that revenue at labs including OpenAI and Anthropic kept growing quickly; S&P Global forecast further growth in data-centre investment into 2026 despite the concerns. Whether the buildout reflected durable demand or an accounting-assisted financing loop remained unresolved, but by December it was a routine subject of business coverage rather than a contrarian position.