1.2.3 Capital vs. Labor Returns

For most of the 20th century, there was an implicit deal. As companies got more productive—as technology improved, as workers learned new skills, as processes got refined—the gains were shared. Not equally, not perfectly, but shared. Productivity went up, and wages went up too. Labor and capital rose together. The boats, if not exactly equal in size, at least floated on the same tide.

In 1980, labor's share of national income was 66%. Workers—through wages, salaries, and benefits—claimed two-thirds of everything the economy produced. Capital—profits, dividends, returns on investment—got the rest. By 2025, labor's share had fallen to 58%, while corporate profits' share of income rose from 7.2% to 11.7%. Do the math: if labor had maintained its 1980 share, American workers would collectively earn an extra $2 trillion per year—an average of $12,000 more for each employed American. That money didn't vanish. It went somewhere. It went to capital.

Tom worked in manufacturing for thirty years. His plant made engine components—good ones—for decent wages. Over those three decades, he watched productivity climb. Robots handled more of the heavy lifting. Computer systems optimized the workflow. The plant could produce twice as much with the same number of workers. Tom's wages, adjusted for inflation, barely budged. When he started in the 1990s, the implicit promise was simple: work hard, help the company succeed, and you'll share in that success. But somewhere along the way, the deal changed. The company succeeded. Productivity soared. Profits hit record levels. Shareholders did great. Tom got a 2% annual raise that barely kept pace with inflation.

This wasn't unique to Tom's plant. It was happening everywhere, and his experience at the machine level was a microcosm of a transformation playing out across the entire economy. AI is now poised to accelerate this shift to a degree not seen since the dawn of industrialization.

For much of the 20th century, labor productivity and median wages moved in close tandem, tracking each other almost precisely from the 1940s through the 1970s and forming the empirical foundation of the post-war economic compact. Then, in the early 1980s, the link snapped. Productivity continued to climb—driven by technology, process innovation, and globalization—but wages decoupled and stagnated. The arrangement that had defined American prosperity simply ceased to function.

Several forces drove this rupture. Declining union membership stripped workers of collective bargaining power. Offshoring gave companies the ability to source labor globally, undercutting domestic wage demands. Shareholder primacy became the dominant corporate philosophy, compelling executives to optimize for stock price rather than for broader stakeholder interests. And as machinery increasingly substituted for human labor rather than complementing it, the returns from productivity gains flowed more readily to equipment owners than to the workers who remained. These forces reinforced one another over decades, steadily eroding the structures that had once ensured workers could capture a share of what they helped create.

AI is not the origin of this dynamic. But it is amplifying it—building on a system already tilted toward capital and accelerating the pace at which gains flow away from those doing the work.

The Numbers Tell the Story

Research tracking AI innovation across European regions found that for every doubling of regional AI deployment, labor's share of income declines by 0.5% to 1.6%. That might sound modest, but compounded over years and across an entire economy, the cumulative effect is substantial. Other research is more direct: AI promotes an increase in the share of capital in the production process, leading to higher returns on capital and exacerbating income inequality. The more AI gets deployed, the more income flows to those who own the technology rather than those who work alongside it.

This shows up clearly in corporate earnings. As of the second quarter of 2025, just eight AI-focused companies—the "AI 8"—accounted for 22% of total S&P 500 earnings. That's eight companies out of five hundred capturing more than a fifth of all profits. OpenAI's annualized revenue hit $13 billion by August 2025. Anthropic reached $7 billion. Nvidia's market cap soared past $3 trillion. The Nasdaq-100 gained 35 percentage points since the start of 2024, driven almost entirely by AI-related demand for chips, cloud computing, and AI services.

Meanwhile, job openings in tech have become scarce and workers' wages are no longer rising rapidly. The boom is real—it's just not reaching the people doing the work.

Where the Money Goes

Between the third quarter of 2020 and the third quarter of 2025, six tech giants—Apple, Alphabet, Microsoft, Oracle, Meta, and Nvidia—spent $1.1 trillion on share buybacks. Across the broader S&P 500, companies returned $1.6 trillion to investors in 2024 alone. Three-fifths of that came through stock buybacks—companies purchasing their own shares to boost prices and reward shareholders. Over the preceding decade, S&P 500 companies had directed 94% of their profits into buybacks and dividends rather than worker compensation, infrastructure, or research and development.

The mechanism is straightforward. Productivity gains flow to the company's bottom line. Profits rise. Those profits get distributed as dividends or used to buy back shares, which increases stock prices. Executives, compensated largely in stock options, get rich. Shareholders, who own the appreciating stock, get richer. Workers who generated the productivity gains in the first place get a cost-of-living adjustment if they're lucky, and a pink slip if they're not. One analysis described the dynamic plainly: firms posted record profits, used those profits to boost stock prices, and simultaneously cut jobs even as earnings climbed. The people who built the product faced mass layoffs while profits generated by their work were redirected to executive compensation and shareholder gains.

This is not an unintended side effect of market forces. It is the predictable outcome of a system organized around shareholder primacy.

The AI Acceleration

Traditional automation replaced manual labor. A robot took over a factory job. The disruption was real for the workers it displaced, but the broader economy could absorb it: service jobs, knowledge jobs, and roles requiring human judgment remained largely out of reach for machines. Those cognitive roles became the new path to the middle class—the jobs workers were told to pursue as manufacturing declined.

AI is now targeting those jobs directly. The accountant, the paralegal, the junior analyst, the customer service representative—these roles required thinking, communication, and problem-solving. They were supposed to be safe. Generative AI can draft legal briefs, analyze financial data, write code, and handle customer interactions at scale, compressing wages and headcount in exactly the occupations workers once fled to after manufacturing automation displaced them.

The economics compound the problem. AI scales fundamentally differently from physical automation. A robot costs money per unit—it must be built, installed, and maintained. AI software, once developed, can be deployed millions of times at near-zero marginal cost. The equation is brutally asymmetric: enormous upfront capital investment, minimal ongoing labor cost. Data centers—the physical infrastructure of AI—represent billions of dollars of investment while providing employment for a relatively small number of highly specialized workers. This is what economists mean when they say AI increases the capital share of production. More of the value creation comes from owning the technology and less from operating it. Capital becomes more productive; labor becomes more dispensable.

The Uneven Picture

There is an important complication to this story: not all workers are losing. Research shows that wages are rising twice as quickly in industries most exposed to AI compared to those least exposed.

Exposure Level Annual Job Growth Inflation-Adjusted Pay Increases
High AI exposure 1.7% 3.8%
Low AI exposure 0.8% 0.7%

Workers in high-AI-exposure roles—those building models, deploying systems, or working in fields where AI augments rather than replaces human judgment—are seeing stronger job growth and meaningfully higher real wages. For those with the right skills and positioning, the AI boom represents a genuine windfall.

But this is a narrow slice of the workforce. Most workers are not AI engineers or data scientists. Most people are in roles that AI will either eliminate or commodify—customer service, administrative support, entry-level analysis, paralegal work. For them, the distribution of gains is starkly unequal. Workers at the top of the skill distribution in AI-adjacent fields are capturing extraordinary returns; everyone else is watching their share of the economic pie shrink. The headline employment numbers may look stable for now, but the character of the work being created—its pay, its security, its prospects for advancement—is diverging sharply from what it replaces.

The Ownership Question

The core issue is not really about AI as a technology. It is about ownership—specifically, who owns the technology and therefore who captures the returns it generates.

For much of the 20th century, workers had meaningful leverage. Unions could bargain. Labor was essential to production. A factory could not run without people, and organized workers could credibly demand a share of the profits they helped create. That leverage was imperfect and contested, but it was real. AI shifts the balance decisively toward capital. Once a model is trained and deployed, relatively few workers are needed to maintain it. Those who are needed—top-tier engineers capable of improving and extending the system—are scarce enough to command high salaries, but too few in number to reshape broader wage dynamics. The result is a winner-takes-all structure: the owners of AI companies and their shareholders accumulate extraordinary wealth, the engineers who build the systems are well compensated, and everyone else participates primarily as consumers paying subscription fees for access to tools that displaced their jobs or compressed their wages.

This need not be the only possible arrangement. Economists and policy analysts have proposed a range of alternatives. Worker cooperatives that share ownership of AI tools across employees would distribute returns more broadly. Public ownership of foundational AI infrastructure—large-scale models or the data centers that run them—could ensure that productivity gains flow to a broader tax base rather than a narrow shareholder class. Universal basic capital proposals would give every citizen an equity stake in AI companies, replicating the shareholder gains that currently accrue only to the wealthy. Each of these approaches faces real political and practical obstacles. But they represent genuine choices. The current arrangement is not the inevitable product of technological progress—it is the result of policy decisions about tax law, corporate governance, antitrust enforcement, and labor rights. Those decisions can be made differently.

The Historical Parallel

This is not the first time a transformative technology concentrated gains at the top before broader pressures forced redistribution. In the early Industrial Revolution, the gains from mechanization flowed overwhelmingly to factory owners. Workers labored in brutal conditions for subsistence wages while capital accumulated staggering wealth. The disparity persisted for decades before labor organizing, political struggle, and sustained policy reform began to shift the distribution. Child labor laws, minimum wages, the eight-hour day, Social Security, the New Deal—each represented a hard-won intervention against a system generating enormous aggregate wealth while distributing it extremely narrowly.

The lesson from that period is not that redistribution is inevitable. Progress was slow, painful, and regularly reversed. Labor clawed back a share of productivity gains only because workers organized, built political power, and sustained pressure across generations. The technology alone did not produce the outcome. Policy choices and collective action did.

The parallel to the present is sobering. The initial wave of AI deployment is benefiting capital overwhelmingly. Historical precedent suggests this could change—but it will not change on its own. The people profiting from the current configuration have no structural incentive to reform it. The $1.1 trillion in buybacks did not happen by accident; it happened because shareholder primacy was the governing norm, executives were compensated to maximize it, and workers lacked the power to demand otherwise. Whether AI becomes a force for broadly shared prosperity or for concentrated enrichment depends on choices that are still being made—about regulation, taxation, labor rights, and the structure of ownership itself.

Summary

The relationship between productivity and wages that defined the post-war economic compact began to break down in the 1980s, and AI is accelerating that rupture. Labor's share of national income fell from 66% in 1980 to 58% by 2025, with the gap representing roughly $2 trillion per year flowing to capital rather than workers. AI deepens this trend by dramatically increasing the capital intensity of production: once developed, AI software can be deployed at near-zero marginal cost, concentrating returns among technology owners while reducing demand for labor. The financial behavior of large technology companies—directing 94% of profits into buybacks and dividends rather than worker compensation over the past decade—illustrates how those gains are being distributed. The picture is not uniform: workers in high-AI-exposure roles are seeing above-average wage and job growth, while those in AI-displaced occupations face compression or elimination. The fundamental issue is one of ownership—who controls the technology determines who benefits from it. History offers both warning and precedent: the Industrial Revolution generated extraordinary wealth while distributing it narrowly, until decades of organizing and policy reform changed the equation. Whether AI produces a different outcome depends not on the technology itself, but on the governance choices, policy interventions, and exercises of collective power that surround it.

Key Takeaways

  • Labor's share of U.S. national income fell from 66% in 1980 to 58% by 2025 — a gap representing roughly $2 trillion per year that flows to capital rather than workers, an average of $12,000 less per employed American annually.
  • AI accelerates this long-running shift: for every doubling of regional AI deployment in Europe, labor's income share falls by 0.5–1.6%; the more AI is deployed, the more income concentrates among those who own the technology.
  • Unlike physical automation, AI software scales at near-zero marginal cost once developed — making it fundamentally more capital-concentrating than previous waves of automation and reducing the leverage workers have to demand a share of productivity gains.
  • The financial behavior of large tech companies illustrates where gains are going: six major firms spent $1.1 trillion on buybacks between 2020 and 2025; S&P 500 companies directed 94% of profits into buybacks and dividends rather than wages, R&D, or capital investment over the past decade.
  • The picture is not uniform: workers in high-AI-exposure roles (engineers, AI-adjacent fields) are seeing job growth and real wage increases averaging 3.8% annually, while workers in AI-displaced roles face stagnation or elimination.
  • The outcome is not technologically inevitable — it reflects specific policy decisions about corporate governance, taxation, antitrust enforcement, and labor rights; alternatives exist (worker cooperatives, public AI infrastructure, universal basic capital) but face real political obstacles.
  • Historical precedent from the Industrial Revolution shows redistribution is possible, but it required decades of organized labor action and policy reform — the technology alone did not produce a more equitable distribution.

Sources:

Last updated: 2026-02-25