The bearish argument that investor Michael Burry — of "Big Short" fame — has made about AI chips is back in circulation, and it centers on an accounting question that critics say the market keeps waving away.
According to coverage from 24/7 Wall St. and Crypto Briefing, the renewed focus is on GPU depreciation: how quickly the expensive graphics processors that power AI systems lose their value, and whether the companies buying them are accounting for that wear honestly.
The core worry is straightforward. If AI chips wear out or become obsolete faster than their owners assume, then those owners may be spreading the cost of the hardware over too many years. That would flatter reported profits today and leave a larger bill hidden in the future — the "GPU math problem" the headlines say won't go away.
The debate got a fresh airing on television. As reported in the coverage, on CNBC's Closing Bell Overtime on the evening of June 30, 2026, a guest said the quiet part out loud about AI hardware, remarking that "the AI memory hardware food chain is still trading as if there's no choice."
The underlying claim is not new — Burry has raised versions of the depreciation concern before — but its revival matters because it strikes at how richly the AI hardware sector is valued. If skeptics are right that useful lifespans are being overstated, some of the earnings underpinning today's lofty chip and memory valuations could prove softer than they look.
Why it matters: with so much of the stock market's recent gains tied to AI chips, a dispute over how long those chips actually last is really a dispute over how much the AI boom is truly worth.