The short version
Sam Altman and Greg Brockman sat down with Ashley Vance and Kylie Robinson for Core Memory this week. Between them, in a single episode, they laid out the entire structural argument for why a foundation like 8GI exists. They did not intend to. They will not agree that they did. But it is on tape, and the clips are thirty minutes apart, and read together they are a thesis.
This post is that thesis.
Quote one. Ten trillionaires, and we prefer that world.
Around the thirty-nine minute mark, Sam lays out what he calls three futures. He only articulates two. Here is the first one in his own words.
"I can see one where, as Greg was saying, the floor comes way up. You know, everybody gets subjectively like 10 times richer. Just the sort of materially abundant prosperity is crazy huge. People are like, 'Man, relative to my life a decade ago, I'm doing great.' But also in that world these people who really learn how to use agents and get a lot of compute together and whatever we have some trillionaires you know maybe 10 trillionaires whatever. So like the floor comes way up like dramatically up but because this is a lever that people can really use, the sort of like most capable most ambitious, the people who already started rich and have access to a lot of compute, inequality gets worse."
Read that again. In the CEO of OpenAI's own mouth, on a major podcast, in 2026: the floor comes way up but because this is a lever that people can really use, the people who already started rich and have access to a lot of compute, inequality gets worse.
Then, a minute later:
"We think it's obvious, Greg, and I think it's obvious that people should prefer the first."
The first. The ten-trillionaire world with worsening inequality. That is the preferred world. Sam concedes the point is "emotionally" uncomfortable for a lot of people. The intellectual argument, in his framing, overrides the emotional one. Accept the ten trillionaires, because the floor still rises.
This is the K-shaped economy said out loud by a person in a position to bend the K back into an 8, and choosing not to.
Quote two. Cheapest possible access. We agree on that.
Three minutes later, Sam says this:
"Kind of no matter what I think everyone should want much more compute, much more infrastructure and the cheapest possible access to AI because otherwise I think you really exacerbate inequality if there's a limited amount of this and the price goes up because of supply and demand and only the rich people have it."
Agreed. This is the correct diagnosis. Cheapest possible access, or inequality compounds. No argument.
Hold on to that sentence. It matters in thirty minutes.
Quote three. Opportunity for everyone, if.
Greg picks up directly after Sam. Listen to the structure of this carefully.
"AI is really opportunity. AI is opportunity for everyone if you have access, right? If you have compute. If you don't have compute, you can't, right? No matter how good you are with agents, if you don't have the compute to run them, you're not going to be able to do much."
Look at the hinges. If you have access. If you have compute. If you don't have compute, you can't.
That is an admission, in plain English, that opportunity under the AI economy is conditional on compute access. Not on intelligence. Not on work ethic. Not on education. Not on the thing a parent is told to teach their child. On compute. And if you don't have compute, you can't.
This is one of the most important sentences said on a tech podcast in the last two years, and almost nobody is going to quote it.
Quote four. Compute is a profit centre.
Thirty minutes later, still the same podcast, Greg says this about the OpenAI business model.
"One thing I think is also worth thinking about is that compute for us is not a cost center, it's a profit center, right, when you deploy it in the product. And so in many ways, our business is extremely simple, right? We rent or buy compute and then we resell it at a margin. And as long as we have some positive margin on it, then it's scalable, right? Because the demand is just unlimited."
Compute is a profit centre. The business is buying compute and reselling it at a margin. The demand is unlimited. As long as the margin is positive, the business scales.
That is an accurate description of the business. I am not disputing the description. I am asking you to notice what happens when you stand it next to Sam's line from thirty minutes earlier.
The four clips, read as one argument.
Put them in order.
- Opportunity in the AI economy is conditional on compute. (Greg)
- The business model of the largest frontier lab is to resell compute at a margin. (Greg)
- Therefore opportunity, in practice, is gated by ability to pay a margin-priced compute bill. (structural)
- The most capable and already-rich use this lever hardest. We get ten trillionaires and worsening inequality. (Sam)
- This is the preferred outcome. (Sam, explicitly)
The argument is self-contained. I did not add anything. I arranged four sentences spoken by two people in a single podcast, and the shape they make is the shape of a caste system with GPUs.
The "cheapest possible access" line from Sam at quote two is the part that does not survive contact with quote four. Cheapest possible access is a cost structure. A margin-priced resale is a cost-plus structure. They are not the same thing. They cannot be the same thing. A company whose scalability depends on positive margin on compute cannot, by construction, deliver compute at cost. The stated goal contradicts the stated business, in the same voice, in the same hour, on the same recording.
This is not hypocrisy. I am not calling anyone a liar. The two men making these statements appear, to me, to be sincere in all four of them. That is the point. The contradiction is structural, not personal. A frontier lab operating as a commercial entity has to resell compute at a margin or it cannot fund the next training run. Everything downstream of that, including "cheapest possible access" and "opportunity for everyone," is rhetoric that the business model cannot cash.
The Spalding Paradox, narrated by its subjects.
I have a name for the shape of this. I have been calling it the Spalding Paradox since last year. The entities with the most commercial interest in minimal AI regulation have become the primary shapers of AI governance. The entities whose margin depends on compute being scarce and expensive have become the primary narrators of what AI is for.
Usually the paradox is something you have to work to prove. You have to line up the lobbying disclosures, the comment letters, the advisory board seats, the regulatory sandbox exits, the revolving door between the labs and the AI offices that were founded to govern them. It is patient work, and it is the work of a policy brief.
The Core Memory episode skipped several steps. The CEO and the President of OpenAI articulated the paradox in their own words. The business, Greg said, is reselling compute at a margin. The outcome, Sam said, is ten trillionaires and worsening inequality. The preferred world, Sam said, is that outcome. And the ethical fig leaf, Sam said, is that everyone should want cheapest possible access, which the business model makes structurally impossible.
This is the paradox, said by its subjects, on a podcast, in 2026. If I were to write an op-ed claiming they had said it, I would have to argue for every step. I don't. The transcript does the work.
The only coherent response.
If opportunity is conditional on compute, and compute at the frontier is resold at a margin, and the margin scales as the frontier does, then any serious intervention on AI inequality has to change one of three things.
One. Change who owns the compute. Sovereign national compute. Public-interest compute pools. State-backed inference capacity made available to citizens the way libraries are made available to readers. This is the infrastructure argument. France has moved. South Africa, in its draft national AI policy, is considering moving. The Irish AI Bill did not. That is a choice Ireland gets to unmake.
Two. Change the pricing structure. Local-first inference. Open weights. Models that run on the device you already own rather than on a rented GPU in someone else's data centre. This is the engineering argument. It is the argument 8gent Code is built around: free and local by default, cloud is opt-in. It is the argument behind Ollama, LM Studio, the open-weights wave from Meta, Mistral, Qwen, and DeepSeek. It is the reason Apple built a Foundation Model into the operating system instead of behind a subscription.
Three. Change the distribution of compute itself. Universal Basic AI. A citizen entitlement, denominated in inference, attached to the same constitutional logic as access to education or healthcare. Not a charity. A floor. The floor Sam described rises for everyone, but only if someone decides to build the floor. Markets do not build floors. States and foundations do.
You do not have to pick one. The honest answer is all three, coordinated, across jurisdictions, with the open-source layer doing the technical work and the state and foundation layers doing the economic and political work. The K becomes the 8 because someone built the curve. Nobody builds it by accident.
What 8GI is doing about it.
I run 8GI because the Core Memory episode is not a surprise to me. It is the clearest possible statement of a position I have been arguing with for two years. The position is that the frontier is private, the margin is real, the outcome is a caste system of compute, and the preferred world, in the mouths of the people who run the frontier, is the one where that system hardens.
I disagree. Not emotionally. Structurally.
The work I do, and the work 8GI does, maps directly onto the three interventions above.
On ownership. 8GI is a nonprofit foundation. Its equity is locked up in the foundation itself, not in a founder or an investor. Its output is Apache 2.0. The compute it funds is meant to be a cost centre, not a profit centre. That is the whole point of the structure.
On pricing. 8gent Code, the agent kernel, runs local by default. The free on-ramp is not a loss leader. It is the architecture. You can run the whole thing on your own machine, against your own model, with no API key and no subscription. We are not competing with OpenAI on price. We are competing with OpenAI on whether price exists at all.
On distribution. The Universal Basic AI argument is not rhetorical. I put it into my Irish AI Bill submission. Lisle Jenneke and I put it into our South African AI Policy submission. It is on the table as a concrete proposal. A floor of AI capability, free at the point of use, denominated in inference, available to every citizen. Public libraries for computation.
None of this is particularly radical. The radical position is the one in the Core Memory clips, which is that a compute-metered caste economy is not only inevitable but preferable to the alternatives. The counter-position, the one 8GI exists to build, is that opportunity conditional on ability to pay is not opportunity. It is rent-seeking with a model card.
The close.
Sam said, on tape, that the ten-trillionaire world is the world that should be preferred. Greg said, on tape, that opportunity is only opportunity if you have compute, and that compute is something his company resells at a margin.
I am not asking you to be outraged by that. I am asking you to take it seriously, because they were. The clips are not out of context. The clips are the context. The thesis they describe is already in motion. The only question is whether the counter-thesis, the one where compute is treated as public infrastructure, arrives in time to matter.
8GI is one answer to that question. There will need to be many answers. If you are a builder, build one of them. If you are a policymaker, write one of them into law. If you are a citizen, refuse to accept that opportunity is something you buy by the token.
The K is a shape. The 8 is a choice. The difference is whether anyone decides to build the curve.
Sources: Core Memory podcast, "Sam Altman and Greg Brockman on 10 years of OpenAI," released April 2026. Timestamps for quoted passages: 39:05–39:42 (Sam, three futures / ten trillionaires), 42:10–42:26 (Sam, cheapest possible access), 42:50–43:10 (Greg, opportunity conditional on compute), 1:00:23 (Greg, compute is a profit centre). Transcript excerpts are verbatim from the show's official transcript. This post is companion to the South African AI Policy submission co-authored with Lisle Jenneke, due 10 June 2026.
By James Spalding · Dublin · 8GI