AI
“OpenAI CEO Sam Altman said it is 'quite terrible' that so much of the world's limited RAM/DRAM supply sits in home computers during the current shortage." (Post also quotes: "From a global allocation perspective, actually, it's quite terrible to have everyone have a bunch of the world's limited supply of DRAM sitting in their own…”
Plain restatementSam Altman, in a recorded interview, described the presence of scarce DRAM in personal computers as a poor global allocation of that resource, using the words "quite terrible," and argued the memory is better utilized in data centers.
Sam Altman did say this. On the Cheeky Pint podcast with Stripe co-founder John Collison, in an episode published on October 29, 2026, he said that from a global allocation perspective it is "quite terrible" for the world's limited DRAM supply to sit in individual people's computers. The post's version of the quote is accurate except that it stops mid-sentence and leaves out the words "and being used 5% of the time," though the post does explain that utilization point in its own words right afterward. One piece of context is missing: the remark came during a discussion about running AI models on your own computer instead of in the cloud, not as a standalone comment about household memory. The surrounding facts the post mentions check out too, since memory prices have climbed steeply in 2026 and OpenAI's own announcement confirms large memory supply deals with Samsung and SK hynix. Nothing found shows Altman proposing any restriction on consumer memory; he made an efficiency argument. The 5% figure is his own estimate, and no published measurement behind it was found.
[drifted from the evidence:] OpenAI CEO Sam Altman [drifted from the evidence:] said it is 'quite terrible' that so much of the [drifted from the evidence:] world's limited RAM/DRAM supply sits in [drifted from the evidence:] home computers [drifted from the evidence:] during the current shortage." (Post also quotes: "From a global allocation [drifted from the evidence:] perspective, actually, it's quite terrible to have everyone have a bunch of [drifted from the evidence:] the world's limited supply of DRAM sitting in their own computer," and adds that [drifted from the evidence:] his argument is utilization, and [drifted from the evidence:] that memory [drifted from the evidence:] prices have risen as AI companies buy supply.)
Sam Altman, [added by the neutral restatement:] in a recorded interview, described the [added by the neutral restatement:] presence of scarce DRAM in [added by the neutral restatement:] personal computers [added by the neutral restatement:] as a [added by the neutral restatement:] poor global allocation of that [added by the neutral restatement:] resource, using the words "quite terrible," and [added by the neutral restatement:] argued the memory [added by the neutral restatement:] is better utilized in data centers.
Red-tinted words in the claim drifted from the evidence. Green-tinted words are what a neutral restatement needs.
The trace / claim to source
- Altman did make the remark, on the Cheeky Pint podcast hosted by John Collison, in episode 39 published 2026-10-29, at roughly the 29 minute mark.
- The words "quite terrible" are his, applied to the arrangement in which the world's limited DRAM supply sits in individual people's computers.
- The post's rendering of the quote matches the published wording closely, allowing for a minor word difference noted below.
- The utilization argument the post attributes to him is his argument. He said the memory in a personal computer is used about 5% of the time and that you want maximum utilization in the cloud in a supply-constrained environment.
- The backdrop is real. DRAM prices rose steeply through 2026 as memory capacity shifted toward AI servers, and OpenAI's own announcement confirms a Samsung and SK hynix target of 900,000 DRAM wafer starts per month for Stargate.
- Describing the moment as a shortage is consistent with his own framing of a "very supply-constrained environment."
- The post's headline sentence and its quotation both stop at "sitting in their own computer," dropping the clause "and being used 5% of the time" that completes the sentence as spoken. The post does restate that utilization point in its next sentence, so a reader does receive it, but the quotation marks enclose a truncated sentence without an ellipsis.
- The remark was made inside a specific exchange about local inference, that is, about people running open-weight AI models on their own machines instead of using data centers. The post presents it as a standalone comment about household RAM during the shortage. The quoted sentence as spoken does not itself name local inference, and several outlets read it as applying broadly, so this is a context gap rather than a reversal of meaning. PC Gamer explicitly notes he was addressing local inference rather than PC gamers.
- The post's closing line, "Where should the world's memory chips go?", is a prompt added by the poster rather than anything Altman proposed. Nothing in the retrieved evidence shows him calling for any policy, restriction, or reallocation of consumer memory. He made an efficiency argument, not a proposal.
- Whether the exact spoken word was "everyone," as the post has it, or "everybody," as Tom's Hardware, Boing Boing, Insider Gaming, Breitbart and others render it. The difference does not change the meaning, and both renderings appear across outlets.
- Whether he said "being used 5% of the time" or "being used maybe 5% of the time." Boing Boing includes "maybe," most others do not.
- The basis for the 5% figure. It is presented in the interview as his characterization of personal computer memory utilization, and no source found gives a measurement or methodology behind it. Treat it as his estimate, not an established statistic.
Altman appeared on Cheeky Pint, the podcast hosted by Stripe co-founder John Collison, in an episode published 2026-10-29. About 29 minutes in, in a discussion of the compute and memory supply chain, the conversation turned to local inference, meaning running open-weight AI models on one's own hardware rather than in the cloud. Altman said that many people assume open-source models running on personal machines mean the world will not need as much data center build-out, and then said that from a global allocation perspective it is "quite terrible" to have everybody holding a share of the world's limited DRAM supply in their own computer while it is used about 5% of the time. He followed with the point that you want maximum utilization in the cloud in a supply-constrained environment, and that many people's intuition runs the opposite way. The quoted sentence is reproduced near-identically by many independent outlets, and the official transcript page for the episode carries the follow-on sentence about maximum utilization. Separately, the backdrop the post describes is documented: DRAM contract and retail prices have risen sharply through 2026 as memory makers shift capacity toward AI servers, with Tom's Hardware tracking a 32GB DDR5 desktop kit at about $450 in mid-October 2026, and OpenAI's own announcement confirming Samsung and SK hynix are targeting 900,000 DRAM wafer starts per month for Stargate.
Complete reasoning
The reply is formatted for pasting into the thread where the claim is circulating.