Who Owns the Future?
AI will not distribute its gains by itself. Ownership, bargaining power, and public rules will decide who benefits from the machines.
We keep asking whether artificial intelligence will make us richer or poorer, as if the machine possesses a temperament. Perhaps it will be generous. Perhaps it will be cruel. Perhaps, after reading enough of the internet, it will simply develop exquisite contempt for everyone involved.
The more useful question is less dramatic and more political: who owns the systems, who sets the terms of their use, and who can claim the value they create? Machines can expand what an economy is capable of producing. They cannot decide how that abundance is divided. That remains a human arrangement, which is unfortunate only because humans are so accomplished at mistaking arrangements for laws of nature.
The Ownership Question
Artificial intelligence arrives dressed as a universal force, but its physical foundations are remarkably concentrated. Training frontier models requires chips, data centers, energy, capital, and specialized talent. According to the United Nations Conference on Trade and Development, just 100 companies accounted for 40 percent of global business research and development spending in 2022. The United States alone hosts roughly a third of the world's top supercomputers and more than half of its computing power.[1]
This does not make broad prosperity impossible. It makes broad prosperity contingent. A tool owned by a few firms and rented to everyone else creates a different economy from one whose gains are dispersed through competition, wages, public investment, and shared infrastructure. The model may be intelligent. The balance sheet is still doing much of the thinking.
Every major technology creates leverage. Ownership decides where that leverage accumulates. Workers may become more productive, consumers may receive cheaper services, and new businesses may appear. Yet none of these outcomes guarantees that the people whose work is reorganized will receive a proportionate share of the gain.
Friction, for Whom?
Economic history has a favorite euphemism for the distance between invention and improvement: friction. The word is pleasantly mechanical. It turns shuttered businesses, weakened bargaining power, and years of insecurity into the faint resistance of two surfaces meeting.
Industrialization raised output on a scale that would once have looked miraculous. It also demonstrated that productive capacity and lived welfare can travel on different timetables. Safer workplaces, shorter hours, social insurance, and meaningful bargaining power did not materialize from the machinery. They were secured through law, organization, and conflict.
The same distinction matters now. UNCTAD's 2025 technology report is unusually direct: technological progress alone does not ensure an equitable distribution of income. Policy, institutions, and domestic capabilities determine whether innovation narrows divides or widens them.[1] Friction is not an argument against technological change. It is a reminder to ask who is being rubbed away.
The Fifteen-Hour Fantasy
In 1930, John Maynard Keynes imagined that technological progress might eventually reduce the working week to about fifteen hours. The forecast was not foolish. Productivity did rise. What failed to arrive was the automatic conversion of productivity into leisure.
That conversion was never a technical problem. It was a distribution problem. A company that saves ten hours of labor can shorten the week, raise wages, lower prices, increase output, reduce headcount, or enlarge its margin. The software does not choose among these possibilities. Management, markets, labor power, and public rules do.
The Organisation for Economic Co-operation and Development estimates that artificial intelligence could add between 0.2 and 1.3 percentage points to annual labor productivity growth in G7 economies, with substantial variation among countries.[2] That is a wide range, but even its upper edge describes capacity rather than destiny.
The labor market evidence is similarly less cinematic than the public argument. A 2025 index produced by the International Labour Organization and Poland's NASK research institute found that one in four jobs worldwide has some exposure to generative AI. Its authors concluded that transformation is more likely than wholesale replacement.[3] Jobs will change unevenly, and the value created inside them will remain open to negotiation. That negotiation is the story.
The Distribution Problem
AI is not entering a neutral economy. It is entering one in which ownership is already exceptionally uneven. The World Inequality Report 2026 estimates that the richest 10 percent own about three quarters of global wealth, while the bottom half owns roughly 2 percent.[4]
This starting point matters. People who own companies, compute, land, and financial assets can absorb disruption while buying exposure to its upside. People who live primarily on wages encounter the same transition as a change to their schedules, bargaining position, and prospects. One group owns the productivity gain. The other is invited to become more productive.
Wealth alone is not the villain of this story, and poverty is not evidence of virtue. The problem is structural asymmetry. When the owners of a technology can set prices, capture rents, and choose the pace of adoption while everyone else negotiates individually, the distributional result has been partially written before the first prompt is typed.
This is why cheerful forecasts about abundance feel both plausible and incomplete. An economy may produce more of nearly everything while access to housing, education, health care, energy, and credit remains governed by scarcity, market power, or political choice. Plenty in the aggregate can coexist quite comfortably with anxiety in the household. It often has.
The Contract We Have Not Written
The sensible response is neither reverence nor panic. It is governance. Competition policy can limit durable concentration. Labor institutions can give workers a voice in how systems are introduced and how gains are shared. Tax policy can distinguish productive investment from rent extraction. Public investment in education, energy, and computing capacity can widen the number of people able to build with the technology rather than merely rent access to it.
Regulation is already moving from abstraction toward obligation. Under the European Union's AI Act, rules for general-purpose AI models began applying in August 2025. Providers must maintain technical documentation, publish summaries of training content, and implement copyright policies. Providers of models carrying systemic risk face additional duties involving evaluation, incident reporting, and cybersecurity.[5] These rules will not settle the distribution question, but they establish something important: technical systems can be made legible to public authority.
The deeper contract remains unwritten. If AI reduces the labor needed to create value, societies will have to decide whether the dividend appears as shorter working lives, stronger public services, broader ownership, cheaper essentials, higher wages, or a taller column of private wealth. Each option is political. None is an emergent property of the model.
There is still time to make the technology ordinary in the best sense: useful, governed, and widely beneficial. But optimism without institutions is only a wager that concentrated power will become generous. History offers thinner support for that proposition than the pitch decks suggest.
The machine does not have an opinion about your circumstances. Its owners do.
Sources
[1] United Nations Conference on Trade and Development, Technology and Innovation Report 2025
[2] OECD Compendium of Productivity Indicators 2025
[4] World Inequality Report 2026, Executive Summary
[5] European Commission, General-Purpose AI Obligations Under the AI Act, 2025