Altman Calls Idle PC Memory "Terrible" as the RAM Shortage Tightens
OpenAI CEO Sam Altman has said it is "terrible" to have RAM sitting in personal PCs rather than in data centers, arguing that the scarce memory is better pooled in the cloud. He made the remarks during an appearance on the Cheeky Pint podcast, according to The Verge, which reported the comments on 30 October 2026.
Altman was asked about supply chain shortages, including the ongoing RAM shortage, and framed the problem in terms of global allocation. He described it as "quite terrible" for the world's limited supply of DRAM (dynamic random-access memory, the working memory that holds data while programs run) to sit inside individual computers, where it is used five percent of the time.
His prescription was consolidation. Describing a very supply-constrained environment for DRAM, he said "you want maximum utilization in the cloud."
The allocation question is already being settled at the supplier level. Samsung and SK Hynix have long-term deals with cloud service operators and with projects including OpenAI's Stargate, and those agreements have reserved much of the world's RAM output over the next few years. Consumer PC and server buyers are competing for the same wafer capacity (the silicon discs on which memory chips are fabricated), and the long-term contracts leave less of it for the open market.
The interview, on a YouTube channel called Cheeky Pint, covered global compute supply problems facing the AI industry more broadly. Altman's comments on personal PCs sat within that wider discussion of how constrained the industry's hardware inputs have become.
The broader context here is the economics of utilization. A DRAM module in a desktop or laptop may spend most of its life idle, since the machine serves one user at a time. A server can time-share the same memory across many workloads, raising the fraction of installed capacity doing useful work. That logic holds as far as it goes, but it applies most directly to hyperscale operators with the scheduling software and workload mix to keep memory busy. It does less for a consumer who needs local memory for latency-sensitive or offline work, and it does not address who gets priority when supply is short.
In my view, the more practical question is whether pooled, cloud-hosted memory can widen access to capacity rather than concentrate it. If consolidation lets smaller developers and researchers rent memory at high utilization, the shortage becomes easier to live with. If long-term supply agreements absorb most of the output, the same shortage raises costs for everyone outside those contracts. The Altman remarks make the trade-off explicit, and the allocation decisions that follow will matter more to working engineers than the framing of any single podcast.
