Sam Altman Calls Running Local AI a Waste of Scarce DRAM Memory
OpenAI's CEO told the Cheeky Pint podcast that running open models on home hardware is a poor use of scarce DRAM that data centers could put to better use.
OpenAI CEO Sam Altman has described running AI models locally as "a terrible allocation of a scarce resource", arguing that the DRAM inside home computers would be better used in data centers. He made the comment on Cheeky Pint, the podcast hosted by Stripe co-founder John Collison, and it quickly spread across social media.
Altman was referring to users who download open models and run them on their own hardware, without relying on anyone else's cloud. In his view, supply is so tight that the priority should be maximum utilization, which only a data center can deliver.
A market short on memory
The remark came during a discussion of compute scarcity. Altman argued that the market is not responding as it should, and that the large memory makers are keeping their investment conservative despite the huge profits of the past twelve months.
The figures back up the tension. Micron, Samsung and SK Hynix control global DRAM production, and they cannot raise output at the pace hyperscalers demand. According to TrendForce, DRAM prices rose between 90% and 95% in the first quarter of 2026, and nothing points to a near-term drop. The three manufacturers have started opening new capacity and even limiting consumer chips to prioritize the HBM memory that data centers need, but building high-performance memory plants is not a quick job.
As a remedy, Altman wants new players in the supply chain. OpenAI is developing its own chips with Broadcom under the name Jalapeño. Custom silicon would still need DRAM, but it allows far more optimization. SpaceX, Amazon, Meta, Microsoft and Google, with its longer TPU experience, are pursuing similar paths.
The other side of the argument
The stance sits awkwardly with OpenAI's own history. In August 2025 the company released gpt-oss, its first open weights since GPT-2, designed to run locally, and in January of that year Altman said OpenAI had been wrong about open source. That was before the current DRAM crunch.
People who run local models have their own reasons: specific needs, offline use, or distrust of how a company like OpenAI handles their data. Open models now solve many everyday tasks, and free cloud tools already cover most needs, so it is unclear how much of a dent this niche really makes in the supply chain.
Apps worth a look
SUMO: the number puzzleTap the right tiles, add them up, hit the target. Easy to learn, hard to master. Free, for all ages.
Dialo: learn languages by speakingTalk to an AI conversation partner; every sentence comes back corrected and explained in your language. 9 languages, A1 to C2.
ediviewer: read EDI files like documentsOpen EDIFACT and X12 messages, see every segment explained, convert to JSON/XML, export PDF and Excel. Runs locally, no upload.
