adtestbench Higgsfield
EN

AI news · OpenAI · memory supply

Altman says scarce memory belongs in data centres

OpenAI’s chief executive argues that running AI on home computers wastes memory chips that would work harder in the cloud.

By Alexandre S. , 02:30 UTC

Sam Altman, OpenAI’s chief executive, says running AI models on home computers is a poor use of the world’s memory chips. In an interview on Cheeky Pint, the Stripe co-founder John Collison’s podcast, published on 29 October 2026, he argued that DRAM, the working memory in every computer, is in short supply and that chips sitting in personal machines are “being used 5% of the time”. The cloud, he said, keeps them busy.

Retro-futurist illustration: a striped sunset over a grid horizon under a starry sky, with a ringed globe standing on the horizon.
Drawn by adtestbench from “OpenAI CEO says it's 'quite terrible' for your PC to have RAM”,

He was answering the hope that open models running locally would reduce the need for data centres. Altman also said on the podcast that making memory had rarely been a good business over long periods until now. Tom’s Hardware points out that OpenAI’s main business is closed models that run in its own data centres.

His case rests on supply: memory is scarce right now, he said, so it should sit where it runs most of the day.

Some cloud GPU prices also rose the same week. On our 30 October check, Runpod’s on-demand serverless rate for an RTX 5090 rose from $1.58 to $1.79 an hour, and the H100 rate from $4.79 to $5.39, while its hourly Pod rates stayed the same.

Runpod’s 30 October rise hit only the serverless tier, where a card can sit idle between requests: a serverless RTX 5090 now costs $1.79 an hour, against $1.19 for the same card rented as a Secure Cloud Pod, which rewards the full-time use Altman argues for.