OpenAI CEO says it's 'quite terrible' for your PC to have RAM
OpenAI CEO Sam Altman has data centers full of memory, but seemingly it's the 32GB of RAM in your desktop that's the problem. In an interview over beers with the YouTube channel Cheeky Pint, Altman said that local inference, and, by extension, people having DRAM in computers at home, is "quite terrible."
Go deeper with TH Premium: AI and data centers
- The data center cooling state of play
- The custom AI ASIC state of play
- America's AI chip rules keep changing - and the rest of the world is paying the price
- GTC 2026: Ian Buck press Q&A transcript - VP of Hyperscale and HPC speaks out on shelving CPX and shipping LPU decode this year
- Demand for data center CPUs has surged, and AI agents are responsible
When discussing local inference on home computers with open-source models, Altman discussed the assumption that people will be able to run local AI on their own computers, the ones they may already own or want to build. Then he explained why that's a bad idea (at the 29-minute mark):
"From a global allocation perspective, actually, it's quite terrible to have everybody have a bunch of the world's limited supply of DRAM sitting in their own computer and being used 5% of the time," Altman said. "You want maximum utilization in the cloud if you're trying to deal with this very supply-constrained environment. And somehow many people's intuition is the opposite of that."
While OpenAI does have some open-source models, its primary business is in closed-source models that run in data centers, so he's really defending the company's business model.
OpenAI isn't exactly threatened by some laptops and gaming rigs just yet, but local AI could mean people one day paying less for subscriptions to models like GPT-6 Luna if they have the power to run them at home. Altman's talk suggests that the company would prefer that more RAM go to OpenAI's data centers rather than home computers, which also sparks enthusiast fears of not being able to own your own hardware or software, but rather rent an agent and a computer in the cloud.
Some of what Altman is saying is based on an idea that data center buildouts won't slow down. It also assumes that RAM being used in AI data centers 24/7 is more worthwhile than RAM that is used for spreadsheets, web browsing, gaming, and maybe even local AI occasionally.
The rest of the wide-ranging interview includes talk of this week's PHP vulnerabilities, sandboxing models in training, and OpenAI's research in mathematics.