Sam Altman's case against local RAM bets everything on the cloud, and users have reason to push back

Sam Altman, CEO of OpenAI, is facing criticism over a remark about personal computers. According to the short video this piece is based on, he said it is inefficient for users to have RAM (the short-term memory a computer uses to run programs) in their own machines. His reasoning is that memory reaches its highest utilization in the cloud, and that pooling it there is how to handle tight supply.

The source does not say where or when he said it, and it gives no figures. We are only working from that summary, so we will not guess at his exact wording.

The narrow argument holds up. Memory in a personal computer sits idle most of the day. In a data centre, the same chips can be shared across many users and kept busy around the clock. When chips are in short supply, putting them where they get used most is a reasonable engineering view.

The problem is who is making the argument. The head of one of the companies that rents out compute in the cloud is saying people should own less compute themselves. That is a conflict of interest, and it explains the backlash.

Our thesis: efficiency is not the only thing that counts. Local hardware buys privacy, offline use, independence from subscriptions and resilience when a provider changes its prices or terms. Those benefits do not show up in a utilization metric.

Our reading: this fits a pattern we keep tracking. The fight in AI is moving toward control of the plumbing: distribution, infrastructure and access. A world where all memory lives in a few cloud providers is a world of more concentrated power. Our main short-term worry is concentration, not scarcity.

We see the long-term picture differently. Open-weight models (whose weights anyone can download) are getting good enough to run on personal hardware. Models under 32 billion parameters, like Qwen and Gemma, already handle a lot of everyday work on a single machine. That is how the abundance we expect actually reaches people: cheaper intelligence that you run and control yourself. Cloud and local computing will coexist, and the healthy outcome keeps both options open. Today's supply crunch should not become an excuse to make dependence on the cloud permanent.