OpenAI chief Sam Altman’s comments on underused computer memory have drawn criticism from PC enthusiasts, highlighting competing priorities between shared AI infrastructure and consumers seeking sufficient capacity for demanding applications, local processing and future upgrades.
OpenAI Chief Executive Officer Sam Altman has prompted debate over the use of memory in personal computers after arguing that shared cloud infrastructure could make more efficient use of scarce supplies than personal machines running occasional workloads.
Speaking on the Cheeky Pint podcast, Altman questioned the resource efficiency of equipping computers with dynamic random-access memory, or DRAM, when that capacity may remain idle for much of the time. His remarks came during a discussion about running artificial intelligence models locally.
Altman described the arrangement as inefficient from a global allocation perspective, suggesting that memory pooled in data centres could serve multiple users and achieve higher utilisation. He characterised personal computer memory as being used only 5% of the time.
The comments drew criticism from PC enthusiasts, who questioned whether continuous utilisation should determine the value of memory purchased for personal use. The disagreement concerns the benefits of shared infrastructure compared with the flexibility and capacity available on an individual computer.
Local AI and memory utilisation
Local inference involves running an AI model on a user’s own hardware rather than sending requests to a remotely operated service. Larger models can require substantial memory, making RAM capacity an important consideration for people choosing computers for these workloads.
Altman’s argument was that shifting inference to personal machines does not eliminate pressure on memory supplies. Instead, it distributes available capacity across devices that may operate less frequently than shared servers. His comments addressed utilisation rather than establishing how much RAM individual consumers need.
For computer users, however, memory requirements depend on the applications they run and the tasks they undertake. Video editing, gaming and simultaneous use of multiple programmes can require additional capacity even if those workloads occupy only part of the day.
Spare memory can also accommodate changing requirements without an immediate hardware upgrade. From that perspective, unused capacity may represent headroom for heavier workloads rather than an unnecessary purchase.
Supply pressures and consumer choice
The debate comes as AI companies expand data centres requiring large quantities of memory alongside processors and other computing components. Growing infrastructure demand has increased concerns about supply constraints and the prices consumers pay for laptops, desktop systems and upgrades.
For PC buyers, the issue extends beyond whether cloud servers use hardware more intensively. Moving workloads online can reduce some local hardware requirements, but also creates reliance on internet access, service availability and the commercial terms of cloud providers.
Local processing offers a different arrangement, allowing supported workloads to run directly on equipment owned by the user. Choosing between these approaches involves considerations beyond utilisation, including cost, access and control over computing resources.
Altman’s remarks therefore highlight a tension between maximising shared infrastructure efficiency and maintaining the computing capacity consumers want available whenever their own workloads require it.
