Michael Burry says Big Tech is understating AI's costs. It's the write-off risk I'm watching.

Michael Burry says the AI boom is costing Big Tech more than its earnings reports suggest. His argument is that the biggest cloud companies are spreading the cost of their AI hardware over too many years, keeping today's expenses down and profits up. If the equipment becomes uneconomic to use sooner than expected, some of that spending could turn into write-downs.

That caught my attention because earlier this week I compared today's AI buildout with the boom in 2000. I was watching for suppliers to get stuck with equipment they couldn't sell. Burry is looking at which companies are buying it and whether the equipment will stay useful long enough to justify the bill.

Seeking Alpha highlighted his warning on September 25. Burry lays out the argument in his "Cassandra Unchained" Substack, where he suggests write-offs could come "perhaps in 2028 or 2029."

The accounting is straightforward. Depreciation spreads the cost of equipment across the years a company expects to use it. A longer useful life means a smaller annual expense and higher reported profit. Burry argues that the cloud companies' schedules are too generous for AI hardware. He says the companies depreciate their AI chips over four to six years, while he puts the replacement cycle at two to three. He estimates that the difference understates depreciation by about $176 billion from 2026 through 2028, TheStreet reported.

A new generation of chips does not automatically make older hardware worthless. Burry's case depends on that equipment remaining economically useful for less time than the companies assume.

The buyer's side of the write-off

This still isn't the fourth warning sign arriving. Cisco's (CSCO) $2.25 billion inventory charge in 2001 was a recorded loss. Burry is forecasting losses that cloud companies could face later. His argument gives me another place to watch for write-downs, but it doesn't establish that they have happened.

Burry points to the aftermath of the telecom buildout, when depreciation exceeded new capital spending. In his comparison, the S&P 500's aggregate net investment stayed negative for 12 straight quarters, from mid-2003 to mid-2006. That's the part that lines up with what I wrote: in a capital cycle, the damage shows up late, well after the peak.

What I'm listening for

I'm not selling on a short-seller's spreadsheet. But I made the point this week that the word to listen for on earnings calls is "digestion," and depreciation belongs right next to it. Micron (MU) reports on September 30, and the hyperscalers report in late October, so we'll hear how they're talking about useful life and capacity soon enough. Our AI Bubble Index was at 62 out of 100, in its "Heated" band, this week, and I've been tracking the credit side of this in our coverage of wider spreads on AI debt. Burry's warning doesn't move that gauge on its own.

For now, this stays where the rest of the list is: a warning sign I'm watching, not a reason to hit sell.

I am not a financial advisor, and nothing here is investment advice.