Altman says local AI inference wastes scarce DRAM compared with datacenters

Sam Altman says it is terrible for the world's limited DRAM supply to sit in personal computers that are used about 5% of the time.

Rohan Paul, in an X post, adds that per token, datacenters win easily, because local models running at batch size 1 leave hardware underused. The reports give no figures for DRAM supply or utilization beyond the 5% estimate attributed to Altman, and no independent data is cited.