Sam Altman Says PC RAM Is Used 5% of the Time: What It Means for Local AI

On a podcast released October 29, OpenAI’s CEO argued that memory sitting in personal computers is a poor use of the world’s scarce DRAM, and that it does more work in cloud data centers. It is an opinion, not a policy, but it lands in the middle of a memory shortage that already shapes mini PC prices.

By StéphanePublished Nov 3, 20263 min read

What Altman said

That local inference is popular, but that DRAM in home computers sits idle most of the day while cloud servers keep it busy.

Speaking with John Collison on the Cheeky Pint podcast, Altman acknowledged that “a lot of people want local inference,” then argued that, from a global allocation view, it is wasteful to keep DRAM in personal computers that use it about 5% of the time. Tom’s Hardware notes that he was not announcing any policy, and that OpenAI’s main business is closed models running in data centers.

Why it matters for memory prices

OpenAI is one of the largest buyers competing with PC makers for DRAM, and memory contract prices are still rising this quarter.

Moneywise reports that OpenAI’s October 2025 letters of intent with Samsung and SK Hynix cover up to 900,000 DRAM wafers a month, “as much as 40% of global output” by a TechInsights estimate. TrendForce expects conventional DRAM contract prices to rise another 10–15% in the fourth quarter of 2026 and warns that PC DRAM supply could decline in 2027. Our 2026 RAM crisis explainer tracks how this reaches retail prices.

The case for local AI

Local models keep prompts and files at home, work offline and have no subscription, which a utilization argument does not address.

Tom’s Guide argues the opposite of Altman, expecting the AI boom to make local hardware more valuable, not less. For privacy, our guide on how private local AI on a mini PC really is covers what stays on the machine.

What it means for buyers

Buy for the models you want to run: about 32GB for 14B models, 64GB for 32B models and 128GB for 70B to 120B models.

My daily PC is a GMKtec K8 Plus with a Ryzen 7 8845HS and Radeon 780M. I tried local AI on it and found it too weak, which is why I would size the memory for the models first and choose the processor second. Because memory is soldered on most AI mini PCs, the capacity you choose at checkout is the one you keep. Our guide to the best mini PCs for local AI lists measured speeds and the memory each model needs.