Sam Altman says DRAM capacity should go to data centers, not personal computers

In an interview on the YouTube channel Cheeky Pint, OpenAI CEO Sam Altman argued that, amid a constrained global DRAM supply, it is inefficient for individuals to keep large amounts of memory in home computers that are used only occasionally. He said pooled memory in data centers can be put to much higher use, contrasting that model with local AI inference on personal machines. The argument favors allocating more memory to cloud infrastructure, where it can serve many users, rather than to PCs that spend much of their time idle. Altman’s position also aligns with OpenAI’s business: its main models run in data centers, while widespread local AI could give users an alternative to paid services.

The argument rests on the assumption that data-center expansion will continue and that high utilization there makes memory more valuable than the same capacity in consumer computers. But PC memory supports far more than AI, including gaming, work, and everyday tasks, and users may value owning hardware even when they do not use it constantly. Local models could also let people avoid subscriptions and reliance on cloud providers, raising concerns about control over software and computing. The comments accompanying the discussion largely framed Altman’s view as serving OpenAI’s commercial interests; some commenters also disputed the claim that data centers make better use of hardware, pointing to reports of underused AI infrastructure.