The Employer's Market: What Monopsony Power Tells Us About AI and Wages

The labor economics concept that best explains what artificial intelligence is doing to wages was named in 1933. Joan Robinson coined "monopsony" — a market with a single dominant buyer of labor — in The Economics of Imperfect Competition, and then waited nearly a century for the data to catch up.

It has now. And the timing, as the AI economy reshapes the bargaining table from below, is worth attending to.

The Mechanism Robinson Named

A perfectly competitive labor market is one where workers have genuine alternatives. When a worker can leave one employer and find comparable work elsewhere, wages must reflect something close to the marginal product of labor. The employer pays roughly what the worker produces, because if they pay less, the worker leaves.

Monopsony describes the departure from this clean model. In markets where a single employer — or a small group of coordinating employers — dominates hiring, workers face a starker choice: accept the offered wage or leave the labor market. The employer captures the difference between what the worker could command in a competitive market and what they actually receive. Robinson called this the "monopsony rent."

For decades, economists accepted that monopsony was theoretically coherent but empirically marginal. The standard response: labor markets are not perfectly competitive, but they're competitive enough that the monopsony rent is small. Estimates from the 1990s and early 2000s suggested labor supply elasticities — the measure of how responsive workers are to wage changes — were high enough to limit employer pricing power meaningfully.

A new wave of research has dismantled that consensus. Arindrajit Dube's The Wage Standard, reviewed this week in Washington Monthly, synthesizes the empirical record: labor supply elasticity is much lower than the prior literature assumed. Workers do not, in fact, respond strongly to wage differences across employers, particularly when credentialing, geography, or occupational specificity constrains their alternatives. The monopsony rent is large, and it is everywhere.

The Economic Policy Institute documented this in a detailed analysis: "In recent decades, it is not technology, but institutional changes — like the decline of unions, the erosion of the federal minimum wage, and a change in macroeconomic policy priorities — that undercut typical workers' leverage and bargaining power in labor markets." The EPI finding is important for what it says about AI specifically: technology is not the root cause of wage stagnation. Unbalanced employer power is. AI does not create this imbalance; it inherits it — and then amplifies it.

What AI Adds to an Already-Unbalanced Market

Here is the specific mechanism that the standard AI-displacement frame misses.

The displacement frame counts jobs lost. It asks: how many workers did AI eliminate? The Goldman Sachs analysis from April found a net negative of approximately 16,000 jobs per month in the US — 25,000 substituted, 9,000 created. Workers displaced face a decade of scarring: a 3% real earnings cut and 10 percentage points of slower earnings growth. This is real and documented.

But there is a second mechanism, less visible, that operates entirely through the threat of displacement rather than displacement itself.

When an employer can credibly threaten to replace part of a worker's job function with an AI system, the bargaining dynamic changes even for workers who are never actually displaced. The employer's outside option — what they can do if the worker demands higher pay — has improved. The worker's outside option has not. The wage that clears the market moves down, not because anyone was fired, but because the relative leverage shifted.

This is the monopsony mechanism applied at the task level rather than the job level. Robinson's insight was that employer market power derives from worker immobility. AI doesn't change that structural immobility — it adds a new source of leverage for employers in markets where that immobility already existed.

The Teaching Sector as Illustration

The clearest contemporary evidence for this mechanism is not in the sectors where AI displacement is most discussed — software, content creation, professional services. It is in public education.

The National Education Association's 2025-26 annual report shows average US teacher salary at $74,495 — up 3.5% nominally from the prior year. The real purchasing-power figure, adjusted for inflation, represents a decline of approximately 5% from 2017. Starting teacher salary nationally averages $48,112. Support staff average $36,360 — a real decline of $2,344 from 2016.

Teaching is a textbook monopsony labor market. Workers are credentialed specifically for this occupation. Geographic mobility is constrained by family, housing, and the specificity of state licensing requirements. A teacher in Panola County, Mississippi faces a market with very few effective employers. The monopsony rent — the gap between competitive wages and the wage actually offered — has been extracting value from teachers for years.

AI has not displaced teachers at scale. But the threat of AI augmentation — adaptive learning platforms, automated grading, AI tutoring systems — has strengthened the hand of school administrators and state legislatures in wage negotiations. "We can do more with fewer staff" is a more credible statement in 2026 than it was in 2016. The threat effect is real even where the displacement is not.

Why This Matters Beyond the Sector Examples

The broader implication is that AI's impact on wages will be systematically underestimated by any analysis that only counts displaced jobs.

A February 2026 Dallas Fed analysis found AI simultaneously aiding and replacing workers, with wage data supporting both trends. This is consistent with the mechanism described above: some workers in AI-augmented roles see productivity gains reflected in wages; many more workers in AI-exposed roles see their bargaining position weaken without being displaced. The net wage effect in the latter category is negative and largely invisible to labor market statistics.

The Princeton AI Snake Oil researchers have made a related point: research on AI and the labor market is still in the first inning. The measurement problem runs deep — standard employment surveys capture displacement poorly and the threat-effect mechanism not at all.

The EPI analysis is most direct on the policy implication: "Like any other technology, AI can be used as a zero-sum tool for increased employer control of work intensity and wages. However, it is the unbalanced power that is the root of this problem — not technological change per se, which could easily boost workers' wages or make jobs easier in more balanced labor markets."

The condition for AI to distribute its productivity gains to workers is the same condition that has been eroding for forty years: worker bargaining power. Robinson named the mechanism in 1933. The AI economy did not create the problem. It is inheriting it, at scale, in a labor market that has spent four decades moving in the wrong direction.

What Would Change This

Three factors constrain the employer's market: unionization rates, minimum wage policy, and macroeconomic tightness. All three are currently in a configuration unfavorable to workers.

Private sector union density remains below 7%. The federal minimum wage has not been raised since 2009; its real value is at a historic low. The Federal Reserve, boxing itself in with inflation from the Iran war energy shock and the tariff structure, cannot cut rates to generate the labor market tightness that was, briefly, boosting worker bargaining power in 2021 and 2022.

The AI economy will not distribute its gains equitably by default. The mechanism that would force equitable distribution — worker leverage — is at its weakest in living memory, precisely as the technology arrives that most advantages the employer side of an already-unbalanced table.

Robinson saw this dynamic in 1933, in the depths of a different technological disruption, in a labor market where power had similarly concentrated on one side of the wage negotiation. She called it by the right name. The current literature is catching up.

Sources: NPR Planet Money, "The Hidden Power Keeping Wages Low," April 21, 2026 | EPI, "Unbalanced labor market power is what makes technology—including AI—threatening to workers," 2026 | Goldman Sachs / Business Insider, AI net negative 16,000 jobs/month with decade-long scarring, April 7, 2026 | NEA Educator Pay and Student Spending Report, 2025-26 | Dallas Fed, AI and labor market dual effects, February 2026 | Washington Monthly review of Arindrajit Dube, "The Wage Standard," April 28, 2026 | PIIE/Brookings, AI and the labor market: still in the first inning, 2026