OpenAI CEO Sam Altman has questioned the efficiency of keeping large amounts of memory in personal computers, arguing that shared cloud infrastructure could make better use of the world's limited DRAM supply.
His comments come as the global memory market faces significant supply constraints, with growing demand from AI data centers contributing to rising prices for consumer RAM and other hardware components.
During a recent podcast discussion with Stripe cofounder John Collison, Altman suggested that memory installed in personal computers often remains underutilized compared with memory deployed in centralized data centers.
He estimated that personal computer memory might be actively used only around 5% of the time, although he did not provide supporting evidence for that figure.
His remarks focused on the efficiency of running AI workloads locally rather than suggesting that consumers should stop purchasing RAM.
Sam Altman Says Cloud Data Centers Can Use Memory More Efficiently
The discussion centered on whether increasingly capable local AI models could reduce dependence on cloud computing infrastructure.
As smaller and more efficient AI models become available, consumers can run certain workloads directly on their computers without sending information to external servers.
However, Altman argued that distributing substantial amounts of memory across individual computers may not be the most efficient use of limited manufacturing resources.
In his view, centralized infrastructure can serve multiple customers and keep expensive computing hardware occupied more consistently.
This approach could improve resource utilization, particularly when demand for memory exceeds available production capacity.
However, the efficiency of cloud computing depends on several factors, including workload demand, server utilization, energy consumption, and network requirements.
Local AI also provides benefits that centralized services cannot always offer, including offline functionality, greater control over personal information, and independence from internet connectivity.
AI Demand Is Contributing to Higher DDR5 RAM Prices
Altman's comments arrive during a period of unusually high memory prices.
AI companies are purchasing substantial quantities of memory for data centers, creating additional competition for manufacturing capacity.
Samsung has reportedly warned that memory shortages could worsen during 2027 and continue affecting the market into 2028.
The company has already committed much of its future high bandwidth memory production to customers through long term supply agreements.
Consumer memory prices have also experienced significant increases in some markets.
For example, certain high capacity DDR5 memory kits in Japan have reportedly increased in price by as much as 761% over the past year.
Separately, one 32GB DDR5 kit purchased for $110 in 2024 was recently advertised for more than $1,500, although cheaper alternatives remain available.
These examples illustrate the severity of pricing increases for particular products, rather than representing uniform increases across the entire memory market.
Memory Shortages Could Continue Through 2027
The shortage may become increasingly difficult to resolve as AI infrastructure investment continues growing.
Industry estimates suggest that memory manufacturers could satisfy only around 60% of worldwide demand by the end of 2027.
Altman also acknowledged that manufacturers have been cautious about expanding production capacity.
Memory companies have experienced significant fluctuations in demand and pricing during previous market cycles, making large investments in new manufacturing facilities financially risky.
Expanding capacity also requires time, meaning additional investment would not immediately resolve existing shortages.
For consumers, the situation raises concerns about the affordability of future PC upgrades.
While shared cloud infrastructure may offer efficiency advantages for certain AI workloads, personal computers continue to require sufficient memory for gaming, productivity, software development, and other everyday tasks.
The discussion highlights an increasingly important question for the technology industry: how to distribute limited memory production capacity between rapidly expanding AI infrastructure and the consumer electronics market.
Discussion (0)
Be the first to comment.