OpenAI CEO Sam Altman believes that people running AI models locally on their computers are contributing to the inefficient use of the world’s limited memory supply. He argues that instead of having large amounts of RAM sitting in personal computers for most of the day, the resources would be better allocated to data centres, where they can be utilised by multiple users.
Altman shared his views during an appearance on the Cheeky Pint podcast, where he discussed the ongoing memory shortage and the growing demand for computing resources. The discussion also touched on the increasing popularity of local AI inference, which allows users to run AI models directly on their own hardware instead of relying on cloud-based services.
Altman Says Cloud Computing Offers Better Utilisation
During the talk, Altman argued that having DRAM installed in individual computers is inefficient from a global resource allocation perspective, particularly when much of that memory is not being used. “It’s quite terrible to have everybody have a bunch of the world’s limited supply of DRAM sitting in their own computer and being used 5% of the time,” he said.
Altman believes that concentrating memory resources in cloud infrastructure would allow them to serve more people and workloads. He added that maximising utilisation in data centres is preferable when dealing with an environment where memory supply is heavily constrained.
However, his argument primarily looks at RAM utilisation from the perspective of running AI workloads. There has been growing interest in running AI models locally, as users look for alternatives to subscription-based AI services and greater control over their own computing resources.
The Lingering Memory Shortage Issue
Altman’s remarks come amid an ongoing global memory shortage, with demand from AI companies and cloud service providers putting considerable pressure on DRAM supply. While data centres may offer better memory utilisation for AI workloads, personal computers still need RAM for everyday applications, gaming and multitasking. There is also the question of whether consumers should have to depend on cloud services to access AI capabilities.
As previously reported, a June forecast by Jefferies Equity Research suggested that memory prices could continue increasing until 2028. It also estimated that AI companies and cloud service providers had already secured around half of global memory supply through long-term agreements, potentially increasing to 70% as more deals are signed. Meanwhile, ADATA chairman Simon Chen previously suggested that the memory shortage could persist for another decade, citing the widening gap between supply and demand as AI adoption continues to expand.
With major AI players continuing to invest heavily in data centres and computing infrastructure, consumers could find themselves competing for an increasingly limited supply of memory components. This could translate into higher prices for RAM modules, PCs, smartphones and other devices that rely on memory chips.
