Altman argues DRAM should go to data centers first during shortages, saying memory idle in gamers' PCs is waste
OpenAI CEO Sam Altman used an appearance on the Cheeky Pint podcast to argue for lowering compute costs, saying that when DRAM supply is tight, concentrating memory in data centers and raising utilization uses scarce resources more effectively.
OpenAI CEO Sam Altman used an appearance on the Cheeky Pint podcast to argue for lowering compute costs, saying that when DRAM supply is tight, concentrating memory in data centers and raising utilization uses scarce resources more effectively. Illustrating with people running local AI models on PCs, he argued many assume open models mean they no longer need data centers, but from a global resource allocation standpoint, if everyone reserved large amounts of scarce DRAM on machines that use it maybe five percent of the time, that is very bad, and cloud resources should be maximized even though most people's intuition is the opposite. The logic has an obvious flaw: most people buy computers not to run local AI inference, since games, professional software, multitasking and the operating system itself all need system memory, so PC memory purchasing and cloud AI procurement are not the same thing and cannot simply substitute for each other. On the industry side, Samsung expects memory supply tightness driven by AI infrastructure to persist through 2028, and gaming hardware prices and availability have tightened, with the company saying at its third-quarter earnings call that near-term capacity additions cannot meaningfully lift supply, so memory cost pressure on consumer electronics will remain in the near term.