On October 29, an episode of Cheeky Pint, the podcast hosted by Stripe co-founder John Collison, went live. Altman talked about the global memory shortage and said something that would later be screenshotted over and over: From the perspective of allocating global resources, it’s pretty terrible for everyone to keep a share of the world’s limited DRAM supply in their own computer. His reasoning was utilization: most of the time, only 5% of the memory in personal computers is in use. To make the most of resources, he argued, it should all be in the cloud.
The remark set off a much bigger firestorm than expected among PC builders. The timing amplified its impact.
That same week, Intel gave fourth-quarter revenue guidance of $15 billion to $16 billion, below market expectations of about $17 billion, and its shares fell more than 20% after hours. The company also expects the PC market to shrink 12% this year and another 10% next year, and says memory supply and prices will not ease next year. HP put it more bluntly: by July, memory accounted for around 35% of the cost of producing a laptop, up from 15% to 18% just one quarter earlier.
Retail prices had already moved. In the U.S., the Xbox Series X costs 60% more than it did at launch in 2020; the PS5 had two price increases this year, and the Switch 2 raised its price in September. Research firms forecast that laptop shipments will fall 13% next year. The extra money ordinary people pay for a memory stick is the real-world context in which that 5% takes effect.
A few days later, Musk posted a meme on X featuring the tavern opening scene from Inglourious Basterds, with just one line: The people deserve to have their RAM too. He didn’t name anyone.
The positioning is highly theatrical. Musk is speaking up for ordinary users, while his Colossus 2 sits in Tennessee, with 550,000 Nvidia GPUs backed by roughly 147 petabytes of high-bandwidth memory, all drawing on the same pool of DRAM wafers. The world’s two biggest memory buyers are joking from each other’s point of view; neither can claim to be above it all.
The real disagreement is about ownership. Altman is talking about an engineering problem: idle memory is wasted. The public hears an allocation problem: who gets to decide what happens to the memory I bought? Engineering problems can be solved through scheduling. Once the allocation question is raised, the person answering it is no longer an engineer.
Altman himself added a caveat. He said memory prices still haven’t corrected themselves, and that the invisible hand is failing right now. That observation is more worth remembering than his 5% estimate. When the invisible hand fails, it means price signals can no longer smoothly redirect capacity back to consumers; long-term contracts and advance payments have already locked the wafers into data centers. At that point, talking about getting resources to where they are most needed amounts to admitting that the power to allocate them has shifted away from prices and toward whoever can secure a long-term contract.
For the secondary market, this clue points primarily to costs. Memory is shifting from a typical cyclical commodity to a strategic material locked up by long-term contracts. That will slow the pass-through of price increases to consumers, but once they arrive, they will hit harder. Upstream cost pressures are squeezing device makers’ margins, and pricing power is moving upstream. The bill for AI capital spending will ultimately show up in ordinary people’s shopping carts as higher price tags on laptops and game consoles.
Musk’s joke spread so quickly because it translated a technical observation into a question of ownership. Taking a memory stick out of a home computer and putting it into a data center would indeed make it more efficient. But even something used only 5% of the time can still be the part many people are most reluctant to replace all year.