Sam Altman Says Local AI Wastes DRAM, Argues for Cloud Inference
OpenAI CEO Sam Altman, speaking on a podcast hosted by Stripe co-founder John Collison, said it is a problem for individuals' PCs to hold massive amounts of DRAM.
1 source 4h ago
Reporting from 1 source: GameBusiness.jp.
Why it matters
Altman's argument against distributed local AI comes from the head of a company whose business runs on closed models in data centers, so the resource-allocation case and the commercial interest point the same way.
Key Facts
- OpenAI CEO Sam Altman made the remarks on a podcast hosted by Stripe co-founder John Collison.
- Altman said DRAM held in individual PCs would be used only about 5% of the time overall.
- Altman argued that concentrating DRAM in data centers and running it nonstop would use the limited supply more efficiently.
- Altman said many people want local inference because they assume it would solve the DRAM supply shortage.
- OpenAI's business centers on closed-source models running in data centers, though it also offers open-source models.
He argued that memory in personal machines sits unused most of the time, while concentrating DRAM in data centers and running it around the clock would use the world's limited supply more efficiently.
Altman framed the question as one of global allocation rather than individual convenience. He said people who want local inference assume it would ease the DRAM supply shortage, but that spreading limited memory across millions of PCs means most of it idles. Inference needs memory for model weights, working data, and key-value caches, and those demands grow with model size and the number of parallel agents.
He did not dismiss local AI outright. Privacy, latency, offline use, and independence from cloud providers remain real advantages, and the number of people running local LLMs is still small. The remark lands as a case for pooling memory in data centers and sharing it among many users.
Written with AI by Yomimono from the 1 cited source below, including Japanese-language reporting where cited. See our editorial policy.
Add Yomimono as a preferred source on Google