Sam Altman says local AI inference wastes scarce DRAM

Sam Altman said on the Cheeky Pint podcast, hosted by John Collison, that local inference is a “terrible allocation” of scarce resources. According to the OpenAI CEO, many people keep DRAM in their own computers, where it is used only 5% of the time, while concentrating it in the cloud would allow greater utilization in data centers. The argument concerns running open models on users’ hardware without relying on the cloud.

Xataka noted that Altman’s position contrasts with users’ interest in privacy and local models. The outlet also cited TrendForce data saying DRAM prices rose between 90% and 95% in the first quarter of 2026, amid limited supply and data centers’ demand for memory.