Altman Says Cloud AI Should Make Better Use of Scarce DRAM

OpenAI chief executive Sam Altman argued that scarce DRAM should be used more intensively in cloud data centers rather than sitting in personal computers, saying local AI use leaves memory underused. His comments, made on the Cheeky Pint podcast, came amid a memory supply crunch driven in part by demand from AI data centers. Altman said people who expect open-source AI models to run on their own computers may overlook the global allocation of limited memory. He described it as “quite terrible” for DRAM to sit in personal computers that are used only 5% of the time, and said cloud computing should aim for maximum utilization in a supply-constrained environment. The 5% figure was Altman’s estimate; he did not provide supporting data or explain how he measured computer use. His broader argument was that concentrating memory in data centers could make more efficient use of a limited resource, rather than distributing it across devices that may sit idle much of the time. Altman also addressed whether memory producers should build more manufacturing capacity to meet demand. He said they should be cautious, pointing to past episodes of overinvestment that left the industry vulnerable...