Altman Sees $100B Memory Cash Flow, Robot-Built OpenAI DCs

OpenAI chief executive Sam Altman said OpenAI is “learning how to build our own data centers”. On Stripe’s Cheeky Pint podcast with John Collison, published on 29 October 2026, he said it has “a robotics program that we’re going to point at making lots of data centers”.

Asked how the compute shortage was going, Altman said: “We always just want more compute”. He also said memory makers, who once would invest $20 billion to make $19 billion, are “each going to make like $100 billion of free cash flow in a year”. OpenAI’s Project Camellia campus in Georgia is described by the Effingham Herald as a $20 billion campus, and OpenAI says it has contracted 3.2 gigawatts of power for it. By our arithmetic, that is about $6.25 billion per gigawatt.

Short explainer · Altman on robots and data centers

A vertical news Short narrated from this piece’s headline and opening line.

Business Pill · Grid Cost Allocation

An evergreen Business Engineer explainer on who pays for the grid behind data centers. It teaches the general idea only and does not discuss OpenAI’s robotics program.

The key insight: Altman tied OpenAI’s answer to the compute shortage to building, not just buying: he said OpenAI is learning to build its own data centers and has a robotics program it will point at making them. By our arithmetic, its Project Camellia campus in Georgia works out to about $6.25 billion per gigawatt.

What Altman Said

Altman listed what OpenAI is doing to prepare for demand: “We made a chip that is an incredible performance at units of intelligence, performance of units of intelligence per watt. We are learning how to build our own data centers.”

He then described the robotics program and what he called self-replicating data centers: “This idea that you can use AI to control robots, and run fabs, and build turbines or solar panels, and build data centers.” He added: “I think we have that within grasp.”

Collison pushed back that data centers are “permitting-bottlenecked and component-bottlenecked”. Altman replied: “We could make more components with robots.” When Collison asked how the market’s solutions to high memory and GPU prices were going, Altman answered: “Bad.”

The memory-maker numbers Sam Altman gave on Stripe's Cheeky Pint podcast, published 29 October 2026: $20 billi
The memory-maker numbers Sam Altman gave on Stripe’s Cheeky Pint podcast, published 29 October 2026: $20 billion of capex to make $19 billion in the old cycle, and about $100 billion of free cash flow a year now, per maker. Altman’s round numbers, not company disclosures.

What Was Posted on X

Wall St Engine on X, 29 Oct, clipping Sam Altman on the compute shortage: “We always just want more compute.” “We are learning how to build our own data centers. We have a robotics program that we’re going to point at making lots of data centers.”

Techmeme on X, 29 Oct: Q&A with Sam Altman (John Collison, Cheeky Pint) on finding PHP bugs, sandboxing superhuman models as an unsolved safety dilemma, robotics scaling data centers, math and agent swarms.

Altman on the Memory Makers

Altman said that for a long time a DRAM producer “would invest $20 billion of CapEx to make $19 billion”, and that now the memory makers are “going to make like $100 billion of free cash flow in a year”, each. He said “the invisible hand is failing right now”, and that memory makers remember “how many times they almost went bankrupt by making too many fabs”.

The memory side of that is in this week’s results. SK hynix said on its 28 October call that DRAM and HBM are effectively sold out through 2029, counting signed contracts plus demand under negotiation, as covered in our SK hynix update. Samsung’s third-quarter deck shows record operating profit, set out in our Samsung piece.

Update (30 Oct): Altman on the cost per gigawatt. In the same Cheeky Pint conversation with John Collison, Altman put a number on the cost he wants to fall. “People give different numbers for the cost per gigawatt measured in the five-year total cost of ownership right now. But let’s say, let’s pick $70 billion as a number,” he said, adding, when Collison asked if that is the cost to install: “And run it for five years.” He went on: “I think if that number were 20 instead of 70, we should be much happier. And we should still demand to get 10x better performance out of it with smarter models every year.” He also said “the fundamental cost of turning the sand into intelligence, that should go down”, and that getting down to $20 billion means energy has to get cheaper and that, of the gross margin stacking between melting sand and getting the token back, “A lot of that’s got to get compressed.” By our arithmetic, $70 billion to $20 billion is a cut of about 71%. Altman called $70 billion an illustrative figure for building and running a gigawatt for five years, so it is not comparable with the $6.25 billion per gigawatt construction figure for Project Camellia above, which covers building only. Wall St Engine also posted the remark on X on 30 Oct.

Update (30 Oct): Altman on what the math cost. In the same Cheeky Pint conversation, Altman said OpenAI “did let it get out of control for one week on the math. Intentionally so,” and that “I bet we could redo Navier-Stokes at 1/100th of the cost a year from when it was done.” Asked by John Collison how much the math cost, he answered: “Order $10 million per… For Navier-Stokes, I think,” and then: “Maybe a little more.” By our arithmetic, 1/100th of $10 million would be about $100,000. Our earlier piece on the run reported a podcast host’s estimate of about $40 million at public list prices, which is a different measure from what OpenAI says it spent.

The Number: $6.25 Billion per Gigawatt

OpenAI’s own post on Project Camellia, of 22 July 2026, says Georgia Power will deliver 3.2 gigawatts in phases between 2028 and 2032, and that OpenAI will pay the full cost of the infrastructure and electric service to serve it. The Effingham Herald describes the campus as a $20 billion project.

By our arithmetic, $20 billion over 3.2 gigawatts is about $6.25 billion per gigawatt, and 3.2 gigawatts over five delivery years is an average of about 0.64 gigawatts a year. The two figures come from different sources: the dollar figure is the Herald’s and the gigawatts are OpenAI’s. The project’s agreements face two lawsuits, as reported in our piece on the Georgia suits.

Update (2 Nov): Altman on pointing OpenAI’s math pipeline at AI research

In the same Cheeky Pint conversation with John Collison (published 29 Oct), Altman said: “I think if you apply the same pipeline that we applied to get our models to be so good at math to AI research, it will probably work.” He added that “at least for a while—maybe for a long while—we are going to have to agree to some limits about how smart we let these single models get.” On what nobody can do yet, he said: “Go discover a new architecture that is as big of a single leap forward as the transformer was. I think that’s where things are heading.” His Navier-Stokes cost bet, “I bet we could redo Navier-Stokes at 1/100th of the cost a year from when it was done,” is a 99% reduction by our arithmetic. The quotes resurfaced on 2 Nov in an X post by @deredleritt3r that speculated on how close OpenAI is to recursive self-improvement; that timeline is the poster’s guess, not Altman’s. For how fast frontier capability is getting cheaper, see Rubin’s tokens per dollar and our Navier-Stokes cost estimate.

The Structural Read

As we read the transcript, Altman named three moves in one answer: a chip measured in intelligence per watt, OpenAI-built data centers, and robots pointed at building more of them.

As we read it, Collison’s pushback, that data centers are permitting- and component-bottlenecked, is the constraint the robotics answer is aimed at; Altman replied that robots could make more components.

By our arithmetic, the 3.2 gigawatts OpenAI says Georgia Power will deliver between 2028 and 2032 average about 0.64 gigawatts a year.

Sam Altman, Cheeky Pint, 29 Oct 2026

“There’s very little, relatively little talk about self-replicating data centers.”

Three Implications

FOR MEMORY MAKERS Altman said the memory makers remember how many times they almost went bankrupt by making too many fabs, which Collison said leads them to be conservative in their capex.

FOR DATA-CENTER BUILDERS Altman said OpenAI is learning to build its own data centers, and OpenAI says it will pay the full cost of the infrastructure and electric service for Project Camellia.

FOR THE COMPUTE SUPPLY Asked how the market’s solutions to high memory and GPU prices were going, Altman answered with one word: bad.

The Business Engineer Lens

This story maps onto the Business Engineer framework The State of AI Data Centers.

The framework puts it this way: “The structural bottleneck isn’t computing, it’s electrons.”

As we read it, Altman’s robotics answer is aimed at a different bottleneck, components and construction, while Project Camellia’s 3.2 gigawatts of contracted power is the electrons side. The framework does not discuss OpenAI’s robotics program.

Related coverage: Suit Targets OpenAI Georgia Deal Carrying $900M Tax Break and Samsung’s Record Profit Took a 9.5 Trillion Won Finance Hit.

What Is Not Established

Altman named no robotics timeline, no data-center count and no budget on the podcast. He did not say which parts of construction the robots would do. We found no OpenAI filing that breaks out its own-built capacity. The $6.25 billion and 0.64 gigawatt figures are our arithmetic from the Herald’s and OpenAI’s numbers.

This Week In The AI Supercycle

Execution Fault Lines, Concentrated Demand, and Geopolitical Firewalls

This week’s issue: the AI build is not short of demand for its parts. It is short of cheap money, and its end revenue rests on a narrow base of large, often venture-funded buyers.

Read the full issue →

Business Engineer Framework

The State of AI Data Centers

A Business Engineer analysis of the power and grid constraints behind the AI data-center build-out.

Read the Map of AI →

The Bottom Line

Altman said OpenAI is learning to build its own data centers and has a robotics program it will point at making them, and said the market is not solving the compute and memory shortage. OpenAI’s Project Camellia campus in Georgia works out to about $6.25 billion per gigawatt, by our arithmetic.

95,000+ executives read Business Engineer for the AI strategy frameworks cited by ChatGPT, Claude, and Perplexity.

A note on sourcing. Altman’s words are quoted from the transcript the Cheeky Pint podcast published on 29 October 2026. Project Camellia figures come from OpenAI’s 22 July 2026 post and the Effingham Herald, as reported in our 28 October piece. Figures marked as ours are our arithmetic. We did not contact OpenAI.

Sources: Cheeky Pint transcript, 29 Oct 2026 · Wall St Engine clip on X, 29 Oct 2026 · FourWeekMBA on Project Camellia, 28 Oct 2026

Scroll to Top
FourWeekMBA