Article · AI translation
IT Home reported on October 31 that OpenAI CEO Sam Altman argued for the need to reduce computing costs in an interview yesterday on the YouTube podcast Cheeky Pint.
He said that, with DRAM memory chips in tight supply, concentrating memory in data centers and increasing its utilization would make more effective use of limited resources.
On the show, Altman discussed the global supply of computing power facing the AI industry and explained his view using the example of running a local AI model on a personal computer.
As open-source models have developed, some users have wanted to run AI models directly on their personal computers to reduce their reliance on cloud services. Altman, however, believes that when memory supply is constrained, concentrating more resources in data centers with higher utilization may better meet overall computing needs.
I think the way supply chains work is actually very counterintuitive. For example, many people favor local AI inference. They assume the problem has now been solved because we have open-source models that can run on our own computers, so there is no need to put all the data in data centers.
But from the standpoint of allocating resources globally, it is actually very bad if everyone ties up large amounts of scarce DRAM in their own computers when that memory may be used only 5% of the time. To deal with such severe supply constraints, cloud resources should be used as fully as possible, but most people's intuition points in exactly the opposite direction.
IT Home notes an obvious flaw in this reasoning. The vast majority of people do not buy computers solely to run AI inference locally: games, professional software, multitasking and the operating system itself all require system memory. Therefore, PC users' reasons for buying memory are not entirely the same as cloud AI service providers' reasons for purchasing it, and their respective DRAM needs cannot simply be treated as interchangeable.
At the industry level, Samsung Electronics expects tight memory chip supplies, driven by AI infrastructure construction, to potentially persist until 2028. This has also led to rising prices and tight supplies in the gaming hardware market.
During Samsung Electronics' third-quarter 2026 earnings call, the company said that intensifying competition to build AI infrastructure and the spread of personal AI agents are expected to continue driving growth in memory demand. Because new capacity is unlikely to significantly increase memory supplies in the short term, tight supply is expected to persist until 2028. This means the pressure of memory costs on consumer electronics is unlikely to ease in the short term.
Related reading:
“Dell CEO warns: Memory shortages will worsen further in 2027”