Sam Altman argues cloud data centers make better use of scarce RAM as DRAM shortage drives PC prices higher
OpenAI CEO Sam Altman has sparked debate by arguing that large amounts of RAM sitting inside personal computers may not be the most efficient use of the world’s limited DRAM supply. His comments arrive at a tense moment for PC buyers, as rising demand from AI data centers continues to push memory prices sharply upward.
Speaking during a recent podcast conversation with Stripe co-founder John Collison, Altman discussed the growing pressure on compute resources and memory production. The conversation touched on the increasing popularity of running open-source AI models locally on personal computers instead of relying entirely on cloud-based services.
Altman said he understands why some people believe powerful local AI models could reduce the need for massive data centers. However, he disagreed with the idea that local computing is the best solution at scale, especially when global DRAM supply is already under heavy strain.
According to Altman, RAM installed in personal computers is often underused. He suggested that it may only be actively used a small percentage of the time, making it less efficient than memory deployed in shared cloud infrastructure, where resources can be allocated to many users more consistently.
His core argument is that cloud data centers can keep expensive and limited memory resources working at much higher utilization rates. In a market where DRAM, HBM, and other memory technologies are in short supply, that efficiency could become increasingly important.
The timing of his comments is significant. The global memory market is experiencing one of its most severe supply crunches in years, largely driven by the explosive growth of artificial intelligence. AI companies require huge amounts of high-bandwidth memory and server-grade DRAM to train and run advanced models, and that demand is affecting the availability and cost of consumer memory.
Samsung has reportedly warned that memory shortages could worsen in 2027 and may continue into 2028. The company’s high-bandwidth memory supply for 2027 is said to already be sold out, with long-term agreements covering a large portion of its production capacity.
Consumers are already feeling the impact. DDR5 RAM prices have climbed dramatically in some markets, especially for high-capacity and high-speed memory kits. In extreme cases, certain DDR5 kits that were affordable in 2024 are now being listed at many times their previous prices, even though cheaper options can still be found.
Industry estimates suggest memory manufacturers may only be able to satisfy around 60% of global demand by the end of 2027 if current trends continue. AI data centers are absorbing a growing share of available production, leaving PC builders, gamers, and workstation users facing higher costs.
Altman also acknowledged that the memory industry has been slow to expand production aggressively. This reluctance is understandable, as memory makers have experienced repeated boom-and-bust cycles in the past. Building new fabrication capacity is extremely expensive, and companies do not want to overinvest if demand later weakens.
Still, Altman’s comments are likely to frustrate PC enthusiasts. For gamers, creators, developers, and professionals, having more RAM locally is not wasteful. It can be essential for multitasking, video editing, 3D rendering, software development, virtualization, and running local AI models.
Local AI also offers advantages that cloud computing cannot always match. Running models on a personal computer can provide better privacy, offline access, lower latency, and independence from subscription-based services. Users who work with sensitive data may prefer keeping everything on their own machines rather than sending information to remote servers.
Altman’s point was not necessarily that consumers should stop buying RAM. His argument focused more on the efficiency of local AI inference compared with centralized cloud infrastructure. Even so, the remarks highlight a growing divide between the AI industry’s need for massive centralized resources and consumers’ desire for affordable, powerful personal hardware.
As AI data centers continue expanding, the competition for DRAM and high-bandwidth memory is unlikely to ease soon. That means PC memory prices may remain elevated, and debates over who should get priority access to limited memory supply will only become more heated.
