Newsletter warns of CPU shortage after GPU and memory squeezes
Gergely Orosz told readers of The Pragmatic Engineer that CPU capacity is the next thing to run short, and that teams needing more compute should secure it now. The public part of his 10 September post carries no pricing and no lead times.
The Pragmatic Engineer newsletter has told compute-heavy teams to reserve capacity ahead of time, on the argument that CPUs are the next component to run short.
The post, The Pulse #191, was published on 10 September 2026 by Gergely Orosz.
Orosz frames the shortage as the third in a sequence. First GPUs, then memory, both driven by demand from AI companies.
His stated cause for the CPU squeeze is AI agents, which he says consume considerably more CPU when they call tools.
The advice attached to it is blunt: if you will need more compute in future, secure it now.
Most of the argument sits behind a paywall. The freely readable portion states the trend and the recommendation, and nothing beyond that.
That matters for what a founder can actually do with it. There are no published figures here on reserved-instance prices, no quoted lead times from any cloud provider, and no named company reporting a delay.
So the claim is a well-sourced practitioner's read on the market, not a measured one. It is worth weighing on Orosz's track record rather than on evidence presented in the post.
For anyone building a 2026 budget around steady CPU spend, the practical question is whether to commit early. Reserved instances trade flexibility for a lower rate and a guaranteed slot; the case for taking that trade gets stronger if capacity really is tightening.
The cheap move is to ask your account manager for current lead times on the instance families you depend on, and to compare a one-year reservation against your on-demand run rate before renewal season.
Neither of those numbers is in the post. Both are on your own invoice.