OpenAI CEO Sam Altman has argued that keeping scarce DRAM in personal computers is inefficient when that memory spends much of its time unused. Speaking during a discussion of the memory shortage, he described that allocation of resources as “quite terrible” and said cloud infrastructure could make better use of the available supply.
The comments came in an October 29 episode of the Cheeky Pint podcast, hosted by John Collison, during a discussion about running AI models locally on users’ computers.
Altman suggested that memory installed in individual PCs might be used only 5% of the time. That figure was part of his argument, rather than a measured utilization rate established in the interview. With supply constrained, he advocated maximizing utilization in the cloud, where shared hardware can serve multiple users.
Memory demand reaches beyond AI
RAM is also essential to everyday computing, gaming and other consumer electronics. Prioritizing utilization in data centers does not remove those needs, while demand from AI infrastructure is adding pressure to an already constrained market.
Samsung’s third-quarter results, published on October 29, described capacity constraints and strong server demand pushing memory prices higher across applications. The company expects AI-driven demand to substantially exceed industry supply in 2027, suggesting that a quick return to easier availability is far from assured.
Gaming hardware buyers face the consequences
The pressure is visible in the wider hardware market. IGN’s Czech report highlighted availability and pricing problems affecting graphics cards, including NVIDIA’s RTX 5080 and RTX 5090, alongside rising console costs.
One striking example is the PlayStation 5 Pro. Tom’s Hardware reported earlier this month that GameStop had listed refurbished consoles at $1,399.99, well above the console’s $899.99 recommended US price following an earlier increase. Those second-hand prices are retailer listings, rather than a new official price from Sony.
Altman’s argument centers on extracting more use from a limited pool of memory. For people looking to upgrade a PC, graphics card or console, however, the immediate issue remains the cost and availability of hardware they need at home.
