Ed Zitron’s case that AI GPUs are piling up in warehouses

AI summary

Why might GPUs be sitting unused? Zitron argues that hyperscalers may have bought GPUs without enough data-center capacity to install and use them. He points to power constraints, construction delays, and shortages of electrical-grade steel, while acknowledging that he does not know the full explanation. This mismatch between purchases and construction is discussed at 00:04:25 and 00:12:56.

How large does he think the warehouse inventory may be? Zitron estimates, with uncertainty, that at least $200 billion worth of GPUs may be sitting in warehouses. What does he allege about Nvidia? Around 00:06:06–00:08:46, Zitron argues that Nvidia may be “channel stuffing” by requiring customers to receive racks before they have a foreseeable way to install them, which he says could inflate Nvidia’s revenue.

What risks does he identify for the AI industry? Zitron says GPUs sitting in warehouses contain high-bandwidth memory, contributing directly to higher RAM prices. He also argues that continued chip purchases make RAM and servers more expensive. He describes hyperscalers as overburdened with these assets and says the AI industry may be close to a breaking point, though the source does not establish what ultimately breaks.

What market or inventory signals is he watching? Zitron identifies hyperscaler capital expenditure as a key indicator: flat spending would be a problem, while higher spending would raise the question of why companies keep buying capacity they cannot use. He also says he is watching token spending and capital expenditure for signs of stress.