June 7, 2026

If AI Changes the Factors of Production, Who Shares in the Gains?

Assiduity AI

The newest AI policy debate is not about model weights, benchmark scores, or chip supply. It is about ownership.

President Trump has reportedly raised the idea of the U.S. government taking a stake in leading artificial intelligence companies. The stated logic is straightforward: if AI creates the next generation of trillion-dollar firms, the American public should share in the upside. The Washington Post reported the proposal in the context of planned White House discussions with technology leaders, while Axios separately described it as a way for Americans to participate in the expected profits of future AI giants. [1] [2]

It is an unusual proposal. It may also not survive contact with the law, markets, politics, or the companies themselves. But the important point is not whether this specific policy is adopted. The important point is that the question has become unavoidable.

If artificial intelligence changes the structure of production, who gets the gains?

That is not a new question. It is one of the oldest questions in political economy. For centuries, economists have described production in terms of three broad factors: land, labor, and capital. Land captures location, natural resources, and physical constraints. Labor captures human effort, skill, judgment, and coordination. Capital captures tools, machinery, infrastructure, financial claims, and ownership.

Each major economic transition changes the relative importance of those factors. It changes who has bargaining power. It changes who can claim the surplus. It changes which social promises remain credible.

In an agrarian economy, land was the central productive asset. Wealth and power followed ownership of territory, agriculture, minerals, ports, and water. In an industrial economy, capital became more important, but labor remained indispensable at scale. Factories required workers. Railroads required workers. Mines, mills, offices, ports, and assembly lines required workers.

The modern services economy changed the composition of labor, but not its centrality. Educated labor retained bargaining power because expertise, judgment, analysis, writing, coordination, and institutional memory were difficult to automate. The college premium, the professional class, and the white-collar middle class all rested on the same assumption: skill could still be exchanged for security. Artificial intelligence challenges that assumption.

It does not eliminate labor. It does not mean all work disappears. It does not mean every AI system substitutes for a human worker rather than augmenting one. New technologies often create new tasks, new firms, and new forms of work. The history of innovation is not a simple story of machines replacing people.

But the distributional question remains. If more production can be carried out through capital-intensive systems, i.e., models, chips, data centers, cloud platforms, energy infrastructure, proprietary software, and distribution networks, then more of the surplus may accrue to the owners of those systems.

That concern is consistent with emerging evidence that AI may be capital-biased. CEPR/VoxEU summarizes research finding a statistically significant negative association between AI patenting intensity and labor’s income share across European regions. The finding should not be overstated. It is regional evidence, not a universal law. But it points in the direction political economy would expect: when the productive asset becomes more capital-intensive, labor’s share may weaken. [3]

The mechanism is not hard to see. In earlier industrial eras, scale usually required large increases in physical labor, managerial layers, distribution networks, and local operating capacity. AI firms scale differently. More revenue can be tied to more compute, more data-center capacity, more model usage, and more software distribution, without a proportional increase in labor.

OpenAI is an imperfect but useful example. Reuters reported that OpenAI’s annualized revenue surpassed 20 billion in 2025, up from 6 billion in 2024, and that revenue growth was attributed mainly to an increase in computing capacity from 0.6 gigawatts to 1.9 gigawatts. The point is not that employees do not matter. They do. Research talent, engineering, product, safety, sales, and operations remain essential. The point is that the scaling variable increasingly looks like capital infrastructure: compute, energy, chips, data centers, and distribution. [4]

That is what capital bias looks like in practice.

The productive engine becomes less a workplace filled with people and more a network of owned assets, contracts, models, and machines.

Labor remains necessary, but its claim on each marginal dollar of output may weaken. That is why the ownership debate matters.

The first warning came from offshoring. Offshoring did not make labor irrelevant. It weakened the bargaining position of domestic labor by giving firms a credible alternative. If a job could move from a high-wage country to a lower-wage country, domestic workers had less power to claim the gains from production. Even the possibility of relocation changed the negotiation. Capital became more mobile. Labor became more exposed.

AI may become a sharper version of the same pattern, especially for white-collar work. Offshoring told workers that someone elsewhere could do the job for less. AI tells workers that, in some cases, no person may be needed at all. That is the political novelty.

Past automation debates often focused on manufacturing, logistics, agriculture, and routine work. Generative AI reaches into analysis, drafting, coding, design, legal review, finance, customer support, research, marketing, administration, and management support. Goldman Sachs Research estimated that generative AI could expose the equivalent of 300 million full-time jobs globally to automation and could automate tasks that account for roughly 25% of U.S. work hours. That estimate should be read as exposure, not unemployment. Exposure means tasks can be affected. It does not mean jobs disappear one-for-one. [5]

The issue is not whether every exposed role disappears. The issue is that AI reaches into the task bundles that once supported the bargaining power of educated labor. It does not merely challenge labor at the bottom of the wage distribution. It challenges the security premium of credentialed cognitive labor.

That matters because white-collar labor is not just another employment category. It is the foundation of the modern tax base, the housing market, professional services, retirement savings, higher education, municipal finance, and much of the political center. These workers are employees, homeowners, voters, parents, investors, taxpayers, and institutional managers. If their bargaining power weakens, the effects do not remain confined to the labor market.

The strongest counterargument is also the most historically grounded: this has happened before. Mechanization, electrification, computers, and the internet all disrupted existing work, but they also created new industries, new jobs, and new forms of productivity. On that view, AI anxiety is another version of an old mistake: assuming that visible job displacement is the whole story while missing the new work that innovation makes possible.

That argument deserves respect. It may also be partly right. A 2026 paper by Golo Henseke, “Generative AI at Work: From Exposure to Adoption across 35 European Countries,” using the 2024 European Working Conditions Survey of more than 36,600 workers, finds that adoption remains uneven and that early generative AI adoption has no detectable effect on worker-reported technology-related task restructuring. That is consistent with a transitional phase in which firms experiment with new tools before reorganizing around them. [6]

But AI has three features that make the adjustment problem harder: speed, breadth, and target. Its capabilities are improving quickly. Its exposure is not limited to one sector or skill tier. And it extends to cognitive and creative tasks that were previously considered the safer side of the labor market. Even if AI ultimately creates new work, the transition may be faster than institutions can absorb and broader than labor markets can smoothly reprice.

Land, too, will be repriced: away from some office clusters and toward power-rich data-center sites, grid-connected industrial property, cooling capacity, fiber routes, and jurisdictions that can permit infrastructure quickly. But labor remains the more difficult question.

For most of the postwar period, the central social bargain in advanced economies was built around labor income. Acquire skills. Sell labor. Earn wages. Buy a home. Save for retirement. Fund public services through taxation. Pass opportunity to the next generation.

That bargain was never perfect. It excluded many people, underpaid many others, and depended on institutions that were often uneven and unfair. But it remained politically powerful because it gave large numbers of people a credible path to the gains of growth. AI tests whether that path remains wide enough.

If the productive system becomes less dependent on labor and more dependent on capital, then wages alone may become a weaker mechanism for distributing prosperity. The economy may grow while many households feel less secure. Productivity may rise while career ladders narrow. Firms may become more valuable while the number of people needed to operate them declines.

A society can survive inequality. It has a harder time surviving irrelevance.

That is where political economy becomes political history. When the balance among land, labor, and capital shifts too far, the result is not only economic inequality. It is a crisis of legitimacy.

Political orders survive unequal outcomes when enough people believe the system remains reciprocal: work is rewarded, sacrifice has meaning, property is legitimate, institutions are accountable, and the future remains open to their children. They become fragile when large groups conclude that the system no longer needs them, no longer rewards them, or no longer recognizes them as full participants in the order it protects.

The French and Russian revolutions are useful here not because they are identical, but because both illustrate what happens when society’s ownership structure loses legitimacy. In France, aristocratic privilege, fiscal crisis, food pressure, and exclusion from political power made the old order intolerable. In Russia, war, land hunger, industrial labor, and class politics turned economic grievance into regime collapse. In both cases, the conflict was not only over income. It was over who had a rightful claim on the productive order. [7] [8]

AI raises a modern version of that question. If the next productive system is increasingly owned by those who control models, compute, infrastructure, data, and distribution, while labor’s claim on income weakens, the political issue will not remain confined to wages. It will become a question of ownership, representation, and legitimacy.

The American Civil War belongs in this discussion for a distinct and deeper reason. The Confederacy defended an economic and political order built on slavery: a system in which human labor was treated as property, and millions of people were denied the most basic claims of liberty, representation, and legal personhood. The seceding states’ own declarations repeatedly identified slavery, slave property, and hostility from non-slaveholding states as central causes of secession. [9]

The Union’s war aims evolved, but the conflict ultimately forced the country to confront the question of whether a republic could survive while tolerating an economic system that excluded an entire class of people from the rights the republic claimed to protect.

These examples are morally and historically distinct. AI is not slavery, feudalism, or revolutionary class conflict. The analogy is not moral equivalence.

The lesson is narrower and more important: economic orders require legitimacy. When wealth, power, rights, and representation become too misaligned, political conflict eventually moves beyond ordinary policy debate. It becomes a struggle over who counts, who benefits, and who has a rightful claim on the future.

AI presents a far less extreme but still consequential version of that legitimacy problem: a productive system that may generate enormous wealth while weakening the ordinary route by which most people participate in that wealth. This is why the public-stake debate matters.

It is not merely a question of whether the government should receive a financial interest in a handful of AI companies. It is an early sign that the existing distributional bargain may not survive the next technological transition unchanged.

If labor becomes less central to production, then labor income may become a less sufficient way to share the gains of growth. If capital owns the productive system, capital will claim more of the reward. And the public’s claim is not purely sentimental. The AI economy did not emerge from private capital alone.

It rests on decades of public and quasi-public infrastructure: defense-funded networking research that helped produce the internet, public universities that trained the technical workforce, publicly funded science and mathematics, legal systems that protect corporate formation and intellectual property, energy infrastructure, capital markets, and a vast public web of human-created text. Large language models were built by private firms, but they were also built on a society that had already produced the networks, institutions, knowledge, and data environments that made private scaling possible.

That does not automatically prove that the state should own equity in AI firms. It does mean the public’s claim on AI’s upside is not absurd.

The public helped create the conditions for the asset. It will ask why all of the asset’s gains should accrue only to private capital.

There are many possible answers.

Public ownership is one. Taxation is another. Sovereign wealth funds, broad-based capital accounts, worker participation, public investment, competition policy, and wider access to ownership are others. Some will be more practical than others. Some will be better suited to the American system than others. Some will fail. But the question will not disappear.

The danger is not that capital becomes wealthy. Capital formation is necessary. Investment funds innovation. Infrastructure requires risk. Companies need returns large enough to justify building hard things. The danger is that labor loses its credible claim on the future.

If people believe they can no longer secure their livelihoods through work, the political system will seek another distribution mechanism. That search may be orderly, institutional, and democratic. Or it may become angry, reactive, and destabilizing. The path depends on whether policymakers, companies, and institutions address the legitimacy problem before it hardens.

This is the first AI policy question: who shares in the gains? But it is not the only one.

A citizen can own a share of an AI company and still be harmed by an AI system that fabricates facts, drifts from instructions, or produces an unreviewable decision. A government can tax AI profits and still deploy systems it cannot explain. A worker can receive a public dividend and still face opaque automation in hiring, lending, benefits, healthcare, education, or law.

Ownership answers one question: who benefits from AI’s upside? It does not answer another question: whether AI systems can be trusted when used.

The ownership question is about political economy. The control question is about institutional trust. The country will need both.

Sources:

[1] The Washington Post reported that President Trump said he was considering taking a government stake in leading artificial intelligence companies.

[2] Axios reported that Trump was exploring U.S. government equity stakes in major AI companies such as OpenAI and Anthropic, framed around public participation in future AI gains.

[3] CEPR/VoxEU summarizes research finding a statistically significant negative association between AI patenting intensity and labor’s income share across European regions, including an estimated 0.5 to 1.6 percentage-point reduction in labor share associated with a doubling of AI patent intensity.

[4] Reuters reported that OpenAI’s annualized revenue surpassed 20 billion in 2025, up from 6 billion in 2024, and that growth was attributed mainly to computing-capacity expansion from 0.6 GW to 1.9 GW.

[5] Goldman Sachs Research estimates that generative AI could expose the equivalent of 300 million full-time jobs to automation globally and potentially automate tasks accounting for 25% of U.S. work hours.

[6] Golo Henseke, “Generative AI at Work: From Exposure to Adoption across 35 European Countries,” uses the 2024 European Working Conditions Survey of more than 36,600 workers and finds uneven adoption, averaging 12% across 35 countries, with no detectable effect of early adoption on worker-reported technology-related task restructuring.

[7] Britannica summarizes the French Revolution’s causes as including political exclusion of the financially powerful bourgeoisie, limited rights and impoverishment among lower social groups, fiscal crisis, and tax pressure.

[8] Britannica summarizes the Russian Revolution against the backdrop of World War I, corruption, food scarcity, economic hardship, and political breakdown.

[9] The American Battlefield Trust’s collection of seceding-state declarations includes the seceding states’ own stated grievances over slavery and slaveholding interests.