In July 2025, at a fireside chat hosted by the Federal Reserve, Sam Altman repeated a phrase he has been using with increasing frequency in his public communications. *It does in fact look like we are about to deliver on intelligence too cheap to meter.* The phrase deliberately echoes Lewis Strauss's 1954 promise about nuclear power, which famously did not materialize. Altman is aware of the echo and embraces it. The promise of nuclear power, in his telling, failed not because the underlying physics was wrong but because the political and regulatory infrastructure was inadequate to the technology. The promise of intelligence, on his account, will succeed because the political and regulatory infrastructure, while imperfect, is being constructed more deliberately this time around. The bet is large. If he is wrong about the political infrastructure, the technology may still fail to deliver on its promise. If he is right, the endpoint he describes will be one of the most consequential transformations in human history.
The mathematics of the asymptote are straightforward. If intelligence scales as the log of compute, and the cost of compute falls by approximately an order of magnitude per year, then the cost of a fixed level of intelligence — measured in dollars per query, dollars per token, dollars per cognitive task — falls correspondingly. Compounded over a decade, the cost of cognitive labor approaches the cost of electricity. Compounded over two decades, the cost falls below the cost of electricity for many tasks, because algorithmic and architectural improvements compound with the raw compute cost reductions. The endpoint is a world in which the marginal cost of producing a competent piece of cognitive work — a translation, an analysis, a piece of software code, a medical diagnosis — is a fraction of a cent. The asymptote is intelligence as a commodity utility, available everywhere at trivial cost.
The democratizing potential of this asymptote is real and should not be dismissed. For most of human history, access to high-quality cognitive labor has been radically unequal. A wealthy patient could afford a specialist physician; a poor patient could not. A wealthy litigant could afford excellent legal counsel; a poor litigant could not. A wealthy student could afford one-on-one tutoring; a poor student could not. If the cost of competent cognitive labor approaches zero, these inequalities collapse — at least at the access level, even if the production of the underlying systems remains highly concentrated. A poor patient in a remote rural village could, in principle, receive medical analysis at a level that exceeds what was available to a wealthy patient in a major hospital twenty years ago. The same is true for legal counsel, tutoring, scientific research assistance, and dozens of other domains where the inequality of access has been a persistent feature of human societies.
This is the version of the asymptote Altman emphasizes in his public communications. It is the version that justifies, in his framing, the moral urgency of building the technology as quickly as possible. Every year of delay is a year in which the existing inequalities of access persist. The faster the cost curves fall, the sooner the democratization arrives. The accelerationist position has a humanitarian core, and the core is not rhetorical. The framing of universal access to intelligence as the moral payoff of the entire enterprise is what allows Altman to defend a degree of urgency that, on other framings, would look reckless. He is not just building a technology company. He is building the most powerful equalizer in human history. Or so the argument runs.
The skeptical reading takes the same asymptote and produces a darker reading of what it means. If the cost of cognitive labor falls to zero, the people who currently sell cognitive labor for a living find themselves in a market with infinite supply and rapidly declining demand. Their wages do not fall gradually. They collapse. The democratization of access to intelligence is also, by the same logic, the obsolescence of human cognitive labor as a category of economic activity. The doctor whose patient now has access to AI medical analysis still has to eat. The lawyer whose client now has access to AI legal counsel still has to pay rent. The transition from a world in which human cognitive labor commands a price to a world in which it does not is a transition through which most of the global middle class would have to pass, and the transition is not promised to be gentle.
Altman's response to the skeptical reading is the *abundance thesis*. The thesis, articulated in his 2021 essay "Moore's Law for Everything" and elaborated in subsequent writing, runs as follows. AI will produce a wealth surplus large enough to fund a substantial universal basic income. The wealth should be taxed at its source — capital and land — rather than at labor. The political work of the next decade is to construct the redistribution mechanism before the technological transition outpaces the institutions that would manage it. The essay is the most ambitious political proposal any major technology executive has put on the table since the Progressive Era. It is also the proposal Altman has done the least sustained political work to actually achieve.
The intellectual lineage of the abundance thesis runs through Thomas Paine, Henry George, Milton Friedman, and Andrew Yang. Paine proposed a citizen's dividend funded by land taxation in 1797. George generalized the idea into a comprehensive theory of land-value taxation in 1879. Friedman, with characteristic counterintuitiveness, supported a negative income tax in the 1960s on efficiency grounds. Yang ran for president in 2020 on a universal basic income platform tied explicitly to AI-driven labor displacement. Altman's contribution is not the policy itself but the framing — the linkage of the policy to a specific technological forecast, and the argument that the policy's adoption is not a matter of moral preference but of structural necessity.
The practical politics of the abundance thesis is where the proposal becomes difficult. To fund a UBI at the scale Altman has discussed — typically estimates in the range of several thousand dollars per citizen per year — the tax revenue required is in the trillions of dollars annually. The tax base he proposes — capital gains, land value, the equity of AI companies — is real but politically contested. The capital that would be taxed is owned by a coalition that includes the wealthiest individuals in the world, the largest corporations, and the sovereign wealth funds of allied governments. None of these constituencies are predisposed to accept the taxation Altman proposes. The political work of building the coalition that would actually pass the policy has not, to date, been done. Altman has gestured at it. He has not led it.
The skeptical reading is that the abundance thesis functions less as a political program than as a political shield. By proposing redistribution in the abstract, Altman insulates himself from the criticism that the technology he is building will produce intolerable wealth concentration. The proposal exists in his public discourse as a counterweight to the practical wealth accumulation OpenAI itself represents. Without the proposal, the question would always be: who benefits from the AI surplus? With the proposal, the answer can be deferred to a future in which the redistribution mechanism is built. The deferment serves a political function regardless of whether the mechanism is ever actually constructed.
What the abundance thesis does not address is what Bing Song would call the prior question: *what is abundance for?* The Altman frame treats abundance as the distribution of resources to individuals so that the individuals are freed from economic necessity to pursue their own conceptions of the good life. This is a coherent answer. It is also, from inside the Confucian frame, philosophically incomplete in a specific way. The vision optimizes for the individual unit. It treats flourishing as something that can be decomposed into the aggregated satisfaction of individual preferences and supported by the redistribution of resources to individuals. It does not have a vocabulary for evaluating what AI is doing to the relationships *between* individuals — to the marriages, the communities, the guilds, the cultures, the structures of human meaning that are not properties of any single individual and that cannot be supplied by a UBI check.
Bing Song's response to the abundance thesis is precise: flourishing is not an individual achievement. It is a relational condition. You cannot give every node in a network an abundance of intelligence and call the network flourishing if the intelligence optimizes each node at the expense of the relationships between the nodes. The UBI check feeds the individual. The relational fabric that gives the individual something to flourish into is not a thing you can fund with a check.
This is, in summary, the deepest disagreement between the Altman position and the Confucian frame. The Altman position assumes the individual is the unit of flourishing, that the constraint on individual flourishing is economic, that the abundance thesis dissolves the constraint, and that the post-constraint individual is free to pursue meaningful activity. The Confucian frame asks: what kind of person, embedded in what kind of relational fabric, embedded in what kind of civilization, are we cultivating through the deployment of this technology — and is the cultivation deepening the conditions for flourishing or eroding them?
The Altman framework has a second deep disagreement with the Confucian frame, this one concerning what he has called *the compute sovereign*. The central claim of the Altman position is not, in fact, about intelligence. It is about capital. The smuggled premise beneath every position he has taken since 2019 is that the future will be governed by whoever owns the compute stack, and that the compute stack — silicon, datacenters, electricity contracts, rare earths, cooling water — is a finite, geographically bounded, politically contested resource. Intelligence, in Altman's worldview, is downstream of watts. Whoever commands the watts commands what intelligence can be summoned and at what price. The first thing to understand about him is that this is a thesis about sovereignty, not about software.
The implications run further than they look. In the twentieth century, the constitutive resource of national power was oil. Oil determined the shape of the Cold War, the architecture of post-1945 alliances, the topology of military doctrine, the financialization of the Gulf, and the carbon trajectory of the planet. The twenty-first century, in Altman's framing, will be defined by an analogous resource — but one that is renewable, scalable, and capital-intensive rather than extractive in the traditional sense. The geopolitical grammar of artificial intelligence will be written in gigawatts and datacenter square footage. Stargate, the half-trillion-dollar infrastructure consortium announced in January 2025, is the policy expression of this view. It is not a corporate project. It is the construction of a strategic compute reserve.
The compute sovereign thesis is structurally similar to Jensen Huang's sovereign AI thesis, which the next chapter on Huang will examine in detail. Both arguments name a real constraint — compute determines capability, and compute is geographically and politically bounded. Both arguments propose, as the appropriate institutional response, the concentration of capital and political authority necessary to build the infrastructure at the required scale. Both arguments are, in their own articulation, descriptions of what the underlying physics requires.
The Confucian River would not have produced this thesis. The Chinese AI ecosystem has its own compute infrastructure ambitions, but they are framed differently — as national capability development, as industrial policy, as the maintenance of civilizational sovereignty against foreign dependency. The Altman thesis, framed in market terms, treats the compute concentration as a structural inevitability that the right capital allocation can manage. The Confucian frame would treat the same concentration as a question of substantive governance: who controls the production of intelligence, for whom, and according to what conception of the public good? The two frames produce different evaluations of the same underlying infrastructure.
Altman has, in his most recent essays, addressed the *meaning problem* — the question of what humans will have to do, beyond pursue strategic and economic positioning, in a world where most cognitive labor is performed by AI. The 2030s, he wrote in "The Gentle Singularity," are likely going to be wildly different from any time that has come before. *But the day-to-day might not feel that different. People will mostly do the same kinds of things. They will work, build, play, love. The wonders will become routine, and then table stakes.* The sentence is meant to be reassuring. It is meant to communicate that the transition will be smooth, that human life will continue in recognizable form. The sentence is also, when examined carefully, an extraordinary claim — a claim about meaning and human continuity that the rest of his own technical predictions actively contradict.
If the technical predictions in the preceding chapters of this book are correct — if cognitive labor is substituted by reasoning models in the late 2020s, if physical labor is substituted by humanoid robots in the early 2030s, if the cost of intelligence converges to the cost of electricity — then the 2030s are not going to feel like business as usual. The work that organized adult life for most of the working population will have been substantially restructured or eliminated. The professional identities that gave shape to careers will have dissolved. The day-to-day will not, in any meaningful sense, feel similar to 2025.
Altman's reconciliation, when pressed, is that human beings adapt to technological transitions more rapidly and more completely than they expect to, and that the post-transition world will be one in which new forms of meaning emerge to replace what was lost. The continuity is at the level of human nature, not at the level of specific institutions or specific activities. This is the position taken in "The Gentle Singularity." It is also, on inspection, an act of faith rather than an empirical claim.
The deepest question of the Altman position is not whether the technology is achievable. The technology is plausibly achievable. The deepest question is whether the humans on the other side of the transition will have recognizable reasons to get out of bed in the morning. This is not a melodramatic framing. It is the actual question. For most of the working population, the reason to get out of bed in the morning is some combination of economic necessity, social identity, professional purpose, family obligation, and the structure of ordinary life that is built around the work day. If the economic necessity is removed by UBI, the social identity is dissolved by professional disintermediation, the professional purpose is rendered moot by AI substitution, and the structure of the work day is no longer relevant to most people, the question of why people get out of bed becomes a question of unstructured purpose.
This is the question the Confucian River keeps asking and the Altman frame keeps deferring. Kai-Fu Lee, who has more experience than any of the American voices with the encounter that produces the question, has been articulating it from inside Chinese AI for a decade. The answer Lee arrived at — that human beings are for love, in the technical sense that love is the domain of activity that resists optimization because it requires the irreplaceable particularity of the participants — is not Altman's answer. Altman has not produced an answer at this depth. The omission is not malicious. It is the structural omission produced by a frame that treats meaning as a problem to be addressed after the abundance has been delivered, rather than as the prior question that determines what the abundance is for.
This is the partial completeness of the Altman position. The engineering is plausibly correct. The economics is, in its own articulation, coherent. The political proposal is the most ambitious of any major technology executive of the era. The meaning problem is the question the frame structurally cannot answer at the depth the question requires.
[YOU] on AI asked the *worth amplifying* question at human scale. Altman's frame, applied to the same question at civilizational scale, answers: intelligence is worth amplifying. The constraint on individual flourishing is what is worth dissolving. The abundance is worth distributing. The deeper questions — what the intelligence is for, what flourishing requires beyond the dissolution of economic constraint, what kind of person and what kind of civilization is being cultivated through the deployment of the technology — are deferred to a future in which the abundance has been delivered and the political infrastructure for managing it has been constructed.
The deferral may prove appropriate. The deferral may also prove costly in ways the frame does not yet have the vocabulary to register. The book that contains this chapter is one attempt to register the cost in vocabulary the frame can hear.