Sam Altman Says Your PC’s RAM Is a “Quite Terrible” Use of Scarce Memory. He Also Sells the Alternative.
Sam Altman has looked at the worst memory shortage in years and found a problem: some of that memory is in your PC.
On the Cheeky Pint podcast published October 29, the OpenAI CEO told Stripe co-founder John Collison that, from a global allocation perspective, it is “quite terrible” for everyone to keep a share of the world’s limited DRAM in their own computers, where he says it gets used just 5% of the time. His conclusion: in a supply-constrained world, you want memory fully utilized in the cloud.
The PC community did not take it well. Before you grab a pitchfork, though, it’s worth reading what he actually said, because the real argument is more specific than the headlines, and the weak spots are more interesting.
Cover image: a satire poster, not an OpenAI ad. Its slogans are not Altman quotes.
What Altman actually said
The comment came during a conversation about why memory makers haven’t flooded the market despite sky-high prices. Altman said the “invisible hand” was failing here, because DRAM producers remember the boom-and-bust years and stay careful about building new fabs.
He then turned to local inference, meaning AI models that run on your own hardware. Many people, he said, assume open-source models running on personal computers solve the supply problem. His point was that this gets it backwards. A chip in a data center serves many people around the clock, while a chip in your desktop sits idle most of the day. You can read the full exchange in the official transcript.
So no, he did not say PCs shouldn’t have RAM, and he did not announce a plan to take yours. He was arguing about where scarce memory does the most work. As XDA points out, this is one interview, not an OpenAI policy.
The shortage he’s talking about is real
Altman is right that memory is tight. TrendForce’s contract price forecasts show how hard this year has hit:
| Quarter | Forecast DRAM contract price change | Source |
|---|---|---|
| Q1 2026 | Conventional DRAM up 90–95%; PC DRAM up over 100% | TrendForce, Feb. 2 |
| Q2 2026 | Conventional DRAM up 58–63% | Tom’s Hardware on TrendForce |
| Q3 2026 | Conventional DRAM up 13–18% | TrendForce, July 3 |
The increases are slowing, but they are still increases on top of increases. TrendForce’s July report credits AI server demand for holding prices up, and that is the part PC builders can’t get past: the customers crowding them out are now telling them their share is the inefficient one.
Where the argument falls apart
The utilization logic works on paper. Shared hardware is busier than private hardware. But it leaves a lot out:
- The 5% figure has no source. Altman gave no study or definition for it, so treat it as a rough line from a podcast, not a measurement.
- Your RAM isn’t only for AI. A PC needs memory for games, video editing, browsers and everything else it does. Moving local AI to the cloud doesn’t remove the need for RAM in the machine.
- Idle isn’t wasted. Memory you own is there the instant you need it, with no subscription, no outage and no queue. Your files and prompts also stay on your machine.
- He isn’t a neutral party. Tom’s Hardware notes that OpenAI’s main business is closed models running in data centers. An argument for putting more memory there is also an argument for his product.
PCGeek’s take
Altman’s math isn’t crazy, and his point about utilization will probably hold up for the biggest AI workloads. But “use is highest in the cloud” and “that’s where the memory should go” are different claims. The first is engineering. The second is a choice about who controls computing, and it deserves more skepticism when it comes from someone who rents that computing out by the month.
Owning your hardware has always been a little inefficient. That’s the deal. You pay for capacity you don’t always use, and in return nobody can reprice it, change the terms or switch it off.
Shopping for memory or storage right now? We check one deal a day on the PCGeek deals page. As an Amazon Associate I earn from qualifying purchases.
Would you trade local hardware for cheaper cloud access, or will Altman have to pry the DIMMs from your hands? Tell us in the comments.
Follow @pcgeekinfo for more PC news.
