(Click on image to enlarge)

Michael Burry’s bear case is based on the premise that the AI hyperscalers are under-depreciating their capital spending in order to keep reported GAAP profits artificially high. There is also the suggestion that much of this CapEx will be malinvestment – as it was during the Dot Com Bubble. Yesterday I did the work on Google (GOOG/GOOGL), Amazon (AMZN) and Facebook (META) to see if his argument holds in their case. What I found suggests he is on to something.

It’s all in the chart above. Let me explain what this chart shows. For the last 7 quarters, I calculated the difference between each of the three above mentioned companies’ CapEx and Depreciation. CapEx is the actual cash expenditure spent on capital equipment. Deprecation is the accounting charge taken to reflect the wear and tear on capital (see “The Perils of Adjusted EBITDA”, November 11). The number for each quarter in the chart reflects the combined difference between CapEx and Depreciation for the three companies for that quarter.

The chart shows that this gap has increased dramatically in recent quarters. It was about $15 billion in 1Q24 and more than $50 billion in 3Q25. Over the long term, CapEx and Depreciation should be the same. Certainly it will diverge in the short term in a situation in which companies are ramping up CapEx which is to be expensed over time. But the increase in the gap is certainly grabbed my attention. These three companies are clearly all in in an arms race with each other to build out the infrastructure for AI. They are making a huge bet – and that bet is not being reflected in their Income Statements to date.

That bet continues to ramp up. In their 3Q25 earnings release, META wrote: “Our current expectation is that capital expenditures dollar growth will be notably larger in 2026 than 2025.” That seemingly innocuous sentence is actually a whopper IMO. META is guiding 2025 CapEx to $70-$72 billion. It was $37 billion in 2024. In other words, CapEx is expected to be $34 billion higher in 2025 than 2024 – and the dollar increase next year is expected to be “notably larger”!

There is certainly reason to be concerned that a lot of this massive investment in building out AI will be malinvestment and there will be a reckoning. That’s what happened during the Dot Com Bust. Companies rushed to build out the infrastructure for the internet. While the internet did ultimately prove revolutionary, much of that CapEx was malinvestment and a lot of companies went bust or had to take huge write downs. While it seems likely that AI will be similarly revolutionary, we may have to endure a similar reckoning first.

More By This Author: