Sam Altman wants it on the record that he is on the side of people who buy memory for their own computers. The OpenAI chief executive says a remark about home PCs holding the world’s scarce DRAM was taken out of context, and that he has long pushed the industry to make more of it. The full conversation shows he is half right. The context changes what he meant. It does not change where his argument sends the chips while the shortage lasts.
He made the defence on X after days of jokes at his expense, and he played along while doing it. A user asked him to rename his account “Ram Altman”. He answered “ok yeah” and changed his display name. Then he explained himself: “i actually do want the world to have tons of memory and people to put as much as they want in their computers, and it is f***ing annoying that this was taken out of context.” He added: “I have been arguing for the world to ramp up memory production for a long time.”
What Altman actually said on Cheeky Pint
The line comes from Cheeky Pint, the podcast hosted by Stripe co-founder John Collison, which publishes full transcripts. The passage reads: “But from a global allocation perspective, actually, it’s quite terrible to have everybody have a bunch of the world’s limited supply of DRAM sitting in their own computer. And, being used […] 5% of the time. You want maximum utilization in the cloud if you’re trying to deal with this very supply constrained environment.”
Most of the coverage started the quote there. What came before it matters. Collison had asked why high memory prices were not pulling new manufacturers into the market. Altman replied that “the invisible hand is failing right now”. For decades, he said, a DRAM maker would “invest $20 billion of CapEx to make $19 billion” and live through booms and busts. Now that each of them stands to make “like $100 billion of free cash flow in a year”, in his estimate, “they all remember their history. How many times they almost went bankrupt by making too many fabs.” Only then did he turn to people who want local inference: running open-source AI models on their own machines, on the assumption that this takes the pressure off data centres. The “quite terrible” line was his rebuttal to that assumption.
Later in the same conversation he says the cost of energy and chips “should go down”, that “if it doesn’t go down, I’ll be very sad”, and that “supply’s got to go way up”. On the narrow point, the defence holds. He was talking about AI inference at home, not gaming rigs or video editing, and he asked for more production in the same episode.
Where the context stops helping
In a market short of supply, an argument about allocation is an argument about who gets the chips. “Maximum utilization in the cloud” sends them to data centres, and OpenAI is one of the largest buyers in that market. Samsung and SK Hynix signed letters of intent to supply memory for OpenAI’s Stargate infrastructure project, a commitment The Register put at around 900,000 DRAM wafer starts a month. The letters are not binding, but they show how much of the market OpenAI intends to buy. Notebookcheck pointed out that the argument also fits OpenAI’s business model: closed models that run in its data centres and are sold by subscription.
Even the episode’s clearest statement of abundance is not Altman’s. The “compute glut” in which “anyone who wants a computer, for whatever reason, can have one” is Collison’s summary of his position. Altman did not answer it directly. He replied in data-centre terms, with the five-year cost of a gigawatt of capacity, which he put at around $70 billion and would like to see fall to 20.
The 5% figure has no source
The number that set off the row comes with no measurement behind it. PC Games Hardware noted that it describes how much of the time memory is busy, not how much a machine needs: an operating system keeps data cached, and programs claim more memory in bursts. A gigabyte that sits idle overnight still has to be there when a game or a local model loads. People also run AI on their own hardware for reasons a utilisation figure cannot capture, such as privacy, control over a model a vendor cannot quietly change, and a one-off cost instead of a subscription.
Elon Musk joined in on X with “The people deserve to have their RAM too!” His own AI venture runs the Colossus data centre in Memphis, as PC Games Hardware pointed out.
The shortage is already in shop prices. On November 2, Raspberry Pi raised the price of the 16GB Raspberry Pi 5 by $80 to $385, citing “a lack of alternative supply for 16GB DRAM”. PC Gamer, whose report carried the X exchange to a wide audience, wrote that tech enthusiasts and gamers have been “priced out of new hardware” by OpenAI and the rest of the AI industry. The Cheeky Pint episode was published on October 29, and Altman’s replies on X followed in the first days of November.
Altman’s fix for all of this is more factories. By his own account, the companies that would have to build them remember nearly going bankrupt the last time someone told them the boom was different. Until they change their minds, memory is already being allocated the way he described: to the buyer who pays the most and keeps it busiest, and that buyer is a data centre, not a desktop.