Key points
- What Sam Altman said about RAM on the podcast
- Where OpenAI's 40% memory figure comes from
- Intel says memory prices are hurting PC sales
OpenAI CEO Sam Altman said in a podcast interview that it was "quite terrible," from a global resource-allocation perspective, for people to keep scarce computer memory in their own machines to run AI locally. After days of criticism, he said on X just after midnight ET on November 4 that his comment had been taken out of context and that he wants people to "put as much as they want in their computers."
Altman made the comment in an October 29 episode of Cheeky Pint, a podcast hosted by Stripe cofounder John Collison. After another user suggested it, he changed his name on X to "Ram Altman," The Verge reported. By the afternoon of November 4, the name was back to Sam Altman.
What Altman said about RAM on the podcast
Collison asked how the market was responding to high prices for memory and graphics chips. Altman answered, "Bad." He said DRAM makers, who produce the working memory used in computers and servers, have a long history of booms and busts, and that they remember nearly going bankrupt.
He then brought up local inference, which means running AI models on your own computer instead of in a data center. Altman said many people assume open-source models will solve the problem. He said that "from a global allocation perspective," the trouble is everyone having "a bunch of the world's limited supply of DRAM sitting in their own computer," used "5% of the time."
Altman said you want "maximum utilization in the cloud" when supply is this tight, and that many people's intuition is the opposite. His argument was that memory gets more use in a data center than in a home computer. He gave no source for the 5% figure in the published transcript. The passage was about local AI, and he didn't mention gaming.
Later in the interview, Altman said the cost of chips and energy should come down. For that to happen, he said, "supply's got to go way up."
How Altman responded to the backlash
Elon Musk posted a meme about the comment on November 1. "The people deserve to have their RAM too!" Musk wrote.
Altman's replies began at 11:59 p.m. ET on November 3. He wrote "i lol'd" under a joke image of RAM sticks, and another user asked him to rename his account Ram Altman. "ok yeah," he answered.
Three minutes later, he added that he wants the world to have "tons of memory." Altman used an expletive to say he was annoyed that the comment had been taken out of context. He also wrote that he has been "arguing for the world to ramp up memory production for a long time."
Where the 40% figure comes from
Much of the coverage says OpenAI could take as much as 40% of the world's DRAM supply. The 40% isn't in the OpenAI, Samsung, or SK announcements of their partnerships. It traces to a calculation Tom's Hardware published in October 2025.
On October 1, 2025, OpenAI signed letters of intent with Samsung Electronics and SK Hynix for OpenAI's Stargate data center project. All three cited 900,000 wafers a month, though they described the figure differently.
OpenAI said the two memory makers were targeting 900,000 DRAM wafer starts a month. Samsung said OpenAI's memory demand was projected to reach up to 900,000 DRAM wafers a month. SK said OpenAI's demand for high-bandwidth memory, the type used with AI chips, would reach up to 900,000 DRAM wafers a month, more than double the industry's capacity for that memory at the time. High-bandwidth memory is made from DRAM wafers, which is why it can be counted that way.
Tom's Hardware compared the 900,000 with about 2.25 million wafers a month, a projection of capacity for all types of DRAM in 2025. That produced the 40%.
The 900,000 is the upper end of projected demand, and the 40% compares it with an estimate of 2025 capacity for all DRAM. SK's release treats the demand as high-bandwidth memory, while OpenAI's and Samsung's don't break it down by type, so the 40% doesn't show how much of the memory used in home computers OpenAI would take.
No timeline was given for reaching that demand, so OpenAI's eventual share would depend on how much capacity exists by then. Neither number measures what OpenAI buys today, and we didn't find a public announcement of a final supply contract at that volume.
What the memory shortage means for PC buyers
Contract prices for conventional DRAM rose 93% to 98% in the first quarter of 2026 from the prior quarter, TrendForce estimated. The research firm said on September 30 that it expected a rise of 10% to 15% in the fourth quarter. TrendForce also said PC memory supply could fall in 2027 as more capacity shifts to servers.
Intel (INTC) told investors on October 29, the day the podcast came out, that customers had begun signaling a PC slowdown because of memory prices and availability. Intel expects industry PC unit sales to fall about 12% this year and is planning for a fall of about 10% in 2027, as we reported on October 30.
Why Altman says memory makers are holding back
Altman told Collison that memory makers are each headed for about $100 billion of free cash flow in a year, and that they still remember their history. When Collison said that history leads them to be conservative with capital spending, Altman answered, "Clearly."
Memory makers also face a separate price-fixing lawsuit
In a separate dispute that didn't come up in the interview, a federal class action accuses Samsung, SK Hynix, and Micron (MU) of throttling memory supply to push prices higher. The defendants say the case repeats one an appeals court threw out in 2022. A hearing on their motion to dismiss is set for November 13, and we're following it in our DRAM lawsuit tracker.
