OpenAI CEO Sam Altman has argued that scarce computer memory is better allocated to data centers than spread across personal computers, where it may spend much of its time unused. Speaking on the Cheeky Pint podcast, he addressed supply chain constraints, including the ongoing RAM shortage, as OpenAI’s own purchases put pressure on global supplies.
Altman made the case while discussing local inference, the use of AI models on people’s own computers. He said many people expect open-source models to let them run AI locally and therefore reduce the need for data center resources. In his view, that expectation overlooks how efficiently the available memory is used when it is distributed among individual machines.
Describing the issue as one of global resource allocation, Altman called it “quite terrible” for people to keep portions of the world’s limited DRAM supply in personal computers that use it only five percent of the time. That utilization figure formed part of his argument for concentrating memory in the cloud, where he said the priority should be getting the greatest possible use from constrained supplies.
His comments presented a different view from the intuition he attributed to many supporters of running AI locally. Rather than treating access to open-source models and personal computing hardware as a solution to the need for data centers, he emphasized the amount of scarce memory those computers would hold. He argued that maximizing cloud utilization is preferable in an environment where supply remains limited.
The remarks come as memory manufacturers Samsung and SK Hynix have committed much of the world’s RAM production for the next several years through long-term agreements. Those deals involve cloud service operators and projects such as OpenAI’s Stargate. OpenAI’s buying spree is also squeezing the memory available globally, placing the company’s demand for chips alongside the shortage Altman was asked to discuss.
The effects extend to consumer devices. The billions that cloud operators and major projects are prepared to spend on chips are contributing to higher prices for laptops, phones and other technology. Against that backdrop, Altman’s argument explains his preference for directing limited memory supplies toward shared computing infrastructure, even as people seeking to run AI on their own machines also need access to RAM.
