Altman Says Scarce Ram Is Better Used In Data Centers
OpenAI’s CEO argues that shared cloud capacity makes better use of scarce memory than PCs running AI models locally, as both outlets report rising device costs.
OpenAI CEO Sam Altman has argued that, amid a global memory shortage, scarce RAM is better used in data centers than left in personal computers running AI models locally. Speaking on the Cheeky Pint podcast, he framed the issue as one of how often that memory gets used. Both Engadget and The Verge report that the shortage is also raising costs for devices that need RAM.
Table of Contents
Altman’s case for cloud computing
Altman was responding to the idea that people could run open-source AI models on their own computers and therefore need less data-center capacity, according to The Verge. He argued that, from a global resource-allocation standpoint, it makes poor use of limited DRAM to place it in individual PCs if it is used only five percent of the time. Shared cloud infrastructure, he said, can keep that memory in use more consistently.
Engadget emphasizes that his remarks concerned local inference—running an AI model on a user’s own machine. His argument was not that every use of RAM belongs in a data center. It was a comparison between local and cloud-based approaches to running AI during a supply shortage.
The supply dispute behind the remarks
The comments come as AI data centers compete for memory that also goes into consumer electronics. The Verge reports that Samsung and SK Hynix have long-term deals with cloud operators and projects including OpenAI’s Stargate, reserving much of the world’s RAM output for the next few years. It connects large spending commitments for chips with higher prices for laptops, phones and other devices.
Engadget makes a more specific claim about OpenAI’s role, saying the company has secured roughly 40 percent of global RAM output through various agreements. The Verge does not give that figure, though it also describes OpenAI’s purchasing as a contributor to pressure on supply.
Why PC users may see it differently
Engadget points out that people buy memory for many purposes besides running AI models. Gaming, livestreaming and high-definition video editing can all require more RAM than ordinary computer use. For those users, memory in a personal machine is not necessarily sitting idle simply because it is not serving cloud-based AI.
The Verge also notes that the question reaches beyond gaming PCs to developers using computers for local AI work. Neither report establishes how often those machines use their memory. Altman’s utilization argument therefore addresses one reason to favor data centers, while the outlets’ accounts of rising device prices show why the competition for RAM matters to people buying and using their own hardware.
Sources
This story was compiled by AI from the reports below. Read the originals for the full details.