Ghost GDP: When Economic Output Rises and the Economy Feels Like a Depression

Executive Summary

In the first quarter of 2026, the U.S. economy posted annualized GDP growth of 3.2% — a figure that, by historical standards, should signal broad-based prosperity. Yet consumer confidence surveys plumbed levels not seen since the 2008 financial crisis. Wage growth for the median worker, adjusted for inflation, turned negative. Retail foot traffic declined for the sixth consecutive month.

Welcome to the era of Ghost GDP.

Ghost GDP describes economic output that registers in national accounts but never circulates through the human economy. It is production without paychecks. Growth without prosperity. A rising tide that lifts the yachts while the rowboats sit on dry sand.

This report examines how AI-driven productivity gains are creating a structural divergence between headline economic indicators and the lived experience of most households. We trace the historical parallel to Britain's Engels Pause (1790-1840), project the widening gap between GDP growth and median household income under plausible displacement scenarios, and argue that traditional economic indicators will systematically mislead policymakers during the most consequential labor market transition in modern history.

For investors, the implications are profound: the companies that look most productive on paper may be selling into a consumer base that is quietly evaporating.

The Measurement That Stopped Measuring

Gross Domestic Product was designed for a world where production and consumption were tightly coupled. A factory produced goods. It paid workers. Workers bought goods. The circular flow of income ensured that output and welfare moved in rough tandem. GDP captured this relationship imperfectly but serviceably for the better part of a century.

AI breaks this coupling.

When a company replaces a team of ten analysts with an AI system, GDP accounting treats the transition as a pure productivity gain. The same output — or more — is produced at lower cost. The company's value-added contribution to GDP may actually increase because its profit margins expand. The national accounts record this as economic growth.

But the ten displaced analysts are no longer earning salaries. Their spending at local restaurants, their mortgage payments, their children's college funds — all of this demand evaporates. The AI system does not eat lunch. It does not buy homes. It does not send children to college.

The output shows up in GDP. The income does not show up in paychecks. The gap between these two figures is Ghost GDP — economic production that is real by every accounting standard but phantasmal in its contribution to human welfare.

This is not a theoretical concern. McKinsey's 2025 analysis of AI adoption across Fortune 500 companies found that firms implementing large-scale AI automation reported average productivity gains of 35-40% in affected divisions. Those same divisions reduced headcount by an average of 28%. The productivity gains flowed almost entirely to operating margins and, subsequently, to shareholders. Wage bills in automated divisions declined by 31% even as output increased.

Multiply this dynamic across the economy, and you get a GDP figure that is increasingly disconnected from the economic reality experienced by the majority of the population.

The Engels Pause: A 50-Year Warning From History

The concept of Ghost GDP has a precise historical analog, and understanding it is essential for anticipating what comes next.

Between 1790 and 1840, Britain underwent the first Industrial Revolution. Textile mills, steam engines, and mechanized production drove an unprecedented surge in economic output. British GDP per capita roughly doubled over this period. By every aggregate measure, the economy was booming.

But wages for the average British worker stagnated — or in many cases declined — for nearly five decades. Economic historians Robert Allen and others have documented what they call the "Engels Pause," named after Friedrich Engels, who observed the disconnect firsthand in Manchester's factories. During this period:

• Real wages for unskilled laborers were flat or declining from 1790 to 1840, even as GDP per capita doubled.
• Life expectancy in industrial cities like Manchester and Liverpool actually fell, dropping to as low as 26 years in the worst slums.
• Child labor surged as families needed multiple incomes to survive despite the economy's aggregate growth.
• Caloric intake among the working class declined measurably, with average height — a proxy for nutrition — dropping by nearly an inch.

The economy was growing. The people in it were getting poorer. GDP was rising. Living standards were falling. The aggregate statistics told a story of triumph. The lived reality was closer to catastrophe.

The Engels Pause lasted approximately 50 years before wages finally began to catch up with productivity — and that catch-up required decades of labor organizing, political reform, universal education, and the emergence of entirely new job categories that the original industrialists never anticipated.

The AI displacement threatens to compress a similar dynamic into a much shorter timeframe, with potentially more severe consequences. The Industrial Revolution displaced workers from specific manual tasks. AI is capable of displacing workers across the full spectrum of cognitive labor — from data entry to legal analysis to software development to medical diagnostics.

And here is the critical difference: during the Engels Pause, the factories still needed human hands. The output could not exist without human labor, even if that labor was poorly compensated. AI systems face no such constraint. The Ghost GDP of the AI era is not merely underpaid labor — it is laborless production.

GDP Growth With Rising Unemployment: The Arithmetic

To understand how GDP can grow while unemployment doubles, consider a simplified but representative scenario.

Assume an economy with 160 million workers producing $28 trillion in GDP. Average output per worker is approximately $175,000. Now introduce AI systems that, over a three-year period, automate tasks representing 15% of total labor hours.

In the conventional economic model, those displaced workers find new employment — perhaps at lower wages, perhaps in new sectors — and the productivity gains diffuse through the economy as lower prices, higher real wages, or both. GDP grows, and so does median income. This is the optimistic scenario that most mainstream economic projections assume.

But consider the alternative: the displacement happens faster than re-employment. AI systems do not merely augment productivity — they substitute for labor in ways that eliminate entire job functions. In this scenario:

• 24 million workers are displaced over three years (15% of 160 million).
• AI systems produce the equivalent output of those 24 million workers at roughly 20% of the labor cost (primarily maintenance, compute, and a smaller number of highly skilled AI engineers).
• Total economic output actually increases because AI systems operate 24/7, make fewer errors, and continuously improve. GDP grows by approximately 4-5% annually.
• Unemployment rises from 4% to 10-12%, as displaced workers compete for a shrinking pool of roles that AI cannot yet perform.
• Median household income declines by 8-15% in real terms as labor market slack suppresses wages across the board — not just in directly automated roles.

The result: GDP grows by 5%. Unemployment doubles. Median income falls. Every headline indicator says the economy is thriving. Every household budget says otherwise.

This is not a paradox. It is simple arithmetic. GDP measures production. It does not measure the distribution of production's benefits. When production is decoupled from human labor, GDP continues to rise even as the human economy contracts.

The Divergence Projections

We modeled three scenarios for the divergence between GDP growth and median household income over the period 2026-2032, calibrated to current AI adoption rates and displacement estimates from the Bureau of Labor Statistics, the OECD, and private-sector analyses.

Scenario 1: Gradual Adoption (Base Case)

AI adoption follows the historical pattern of previous general-purpose technologies, with a 10-15 year diffusion curve. Displacement is moderate and partially offset by new job creation in AI-adjacent fields.

• GDP growth: 2.5-3.0% annually
• Median household income growth: 0.5-1.0% annually (real)
• Divergence by 2032: GDP is 18-22% higher than 2026 levels; median income is 4-7% higher
• Ghost GDP gap: ~15% of total GDP growth accrues to capital rather than labor

Scenario 2: Accelerated Displacement (Stress Case)

AI capabilities advance faster than expected — consistent with the pace observed between 2024 and 2026. Corporate adoption accelerates as competitive pressure forces rapid automation. Re-employment rates for displaced workers fall below historical averages.

• GDP growth: 3.5-5.0% annually
• Median household income growth: -1.5% to -3.0% annually (real)
• Divergence by 2032: GDP is 28-35% higher; median income is 8-16% lower
• Ghost GDP gap: ~40-50% of total GDP growth is Ghost GDP
• Unemployment: peaks at 9-12% by 2030 before partially recovering

Scenario 3: Structural Break (Tail Risk)

AI achieves broad cognitive task automation, displacing knowledge workers across law, finance, medicine, software, and management. The displacement is too fast for institutional adaptation.

• GDP growth: 5.0-8.0% annually (driven by AI productivity)
• Median household income growth: -5.0% to -10.0% annually (real)
• Divergence by 2032: GDP is 40-60% higher; median income is 25-40% lower
• Ghost GDP gap: 60-75% of GDP growth is Ghost GDP
• Unemployment: peaks at 15-20%, with labor force participation declining as discouraged workers exit

Even under the base case, the divergence between GDP and median income is the largest since the Engels Pause. Under the stress and tail-risk scenarios, the divergence is historically unprecedented in its speed and scale.

For a detailed analysis of how this income decline translates into reduced consumer spending, see our report on the consumer spending cliff.

Why Policymakers Will Be Flying Blind

The Ghost GDP problem is fundamentally a measurement problem, and it will cause policymakers to make systematically wrong decisions during the critical transition period.

The Fed's Dilemma

The Federal Reserve's dual mandate is maximum employment and stable prices. AI displacement creates a scenario where both indicators send misleading signals:

• Inflation may fall — not because the economy is healthy, but because consumer demand is collapsing as displaced workers cut spending. The Fed could interpret falling inflation as a sign that the economy has room for stimulus, when in reality it reflects a demand crisis.
• Unemployment statistics may undercount the problem because discouraged workers who stop looking for jobs exit the labor force entirely, reducing the headline unemployment rate. The U-3 rate could read 6% while the actual share of working-age adults with full-time employment drops by 10 percentage points.
• GDP growth remains strong, suggesting no recession — even as the median household is experiencing one.

A Fed that relies on these traditional indicators could keep policy too tight (because GDP looks fine) or too loose (because inflation is falling) — neither response addressing the actual structural crisis.

Fiscal Policy Mismatch

Congressional budget projections rely heavily on GDP growth assumptions. If GDP grows at 4%, projections will show rising tax revenues, declining deficit-to-GDP ratios, and reduced urgency for fiscal intervention. The models will say the economy is healthy.

But if that GDP growth is Ghost GDP — accruing primarily to capital owners and corporate profits — the revenue picture may look very different. Corporate tax avoidance, capital gains deferral, and the concentration of income among top earners (who save rather than spend) could mean that nominal GDP growth translates into far less tax revenue than historical relationships would predict.

Meanwhile, the demand for government services — unemployment insurance, food assistance, healthcare subsidies, retraining programs — will surge. The budget math will not work, but the traditional indicators will say it should.

The policy response gap we have documented explores this institutional failure mode in detail.

The BLS Blind Spot

The Bureau of Labor Statistics was designed to measure a labor market where most adults work, most work is full-time, and most workers are employees of firms. AI displacement challenges every one of these assumptions:

• Gig and contract work may surge as displaced workers take whatever is available — but these arrangements are poorly captured in establishment surveys.
• Underemployment — a software engineer driving for a rideshare company — is not reflected in unemployment statistics.
• Quality of employment declines are invisible to headline indicators. A job is a job, whether it pays $150,000 with benefits or $35,000 without.

The statistical infrastructure was built for a different economy. It will continue to report on that economy even as it ceases to exist.

The Consumption Paradox: Production Without Buyers

Ghost GDP creates a terminal contradiction within the market economy: AI systems can produce goods and services at unprecedented scale, but they cannot consume them.

This is not merely a distribution problem — it is an existential challenge to the price mechanism itself. Consider the dynamics that have already begun to emerge in several sectors:

Content and Media: AI can generate articles, videos, music, and images at near-zero marginal cost. Output in these sectors — measured by volume — is exploding. But the humans who previously earned livings creating this content are being displaced, reducing the consumer base that pays for media subscriptions, advertising-supported content, and creative services. More content, fewer paying consumers. Output rises; the market shrinks. For a deeper analysis of the price dynamics, see our report on the pricing death spiral.

Professional Services: AI legal assistants, AI financial advisors, AI diagnostic tools — each displaces well-compensated professionals while making the services cheaper. GDP may register the same output at lower cost (a productivity gain). But the displaced lawyers, financial advisors, and diagnosticians were also the consumers of premium goods and services — the restaurant meals, the vacations, the home renovations that constitute a substantial portion of consumer spending.

Software and Technology: AI coding assistants are already reducing the number of developers needed for a given project. Software output increases. Developer employment — and the substantial spending power of six-figure tech salaries — decreases.

In each case, the production side of the economy looks excellent. The consumption side is quietly eroding. Ghost GDP accumulates in corporate balance sheets, in stock buybacks, in the offshore accounts of the capital-owning class. It does not circulate.

The question that traditional economics has no good answer for: what happens when the economy can produce everything but no one can afford to buy it?

The Measurement Problem: Accounting for an Inhuman Economy

If GDP is no longer a reliable proxy for economic welfare, what should replace it — or at least supplement it?

Several alternative metrics deserve consideration:

Median Income Growth: Unlike GDP, median income directly measures the economic experience of the typical household. A sustained divergence between GDP growth and median income growth is the clearest signal of Ghost GDP accumulation.

Labor Share of Income: This metric tracks the percentage of national income that goes to workers versus capital owners. It has been declining since the 1970s, but AI threatens to accelerate the decline dramatically. A labor share falling below 50% — historically unprecedented in the modern era — should trigger automatic policy responses.

Effective Consumer Base: A new metric is needed to measure the number of households with sufficient disposable income to participate meaningfully in consumer markets. GDP can grow while the effective consumer base shrinks — and this shrinkage will eventually undermine the GDP growth that caused it.

Velocity-Adjusted Output: Traditional GDP treats a dollar of output the same whether it circulates through the economy ten times (as wages spent at local businesses) or sits in a corporate treasury. A velocity-adjusted measure would discount output that does not circulate, providing a more accurate picture of economic vitality.

Time-to-Reemployment: For displaced workers, the duration of unemployment is more economically significant than the headline unemployment rate. If displaced workers take 18-24 months to find new employment (versus the historical average of 6-9 months), the cumulative economic damage is far greater than the unemployment rate alone would suggest.

None of these metrics exist in the current policy framework with the authority and visibility of GDP. Building them is not a technical challenge — it is a political one. And the political system has little incentive to adopt metrics that would reveal the inadequacy of its response.

Investment Implications

For investors, Ghost GDP creates a distinctive risk profile:

Short-term (2026-2028): Companies leading AI adoption will report exceptional productivity gains and margin expansion. Earnings growth will outpace revenue growth — a classic sign of efficiency-driven profitability. Equity markets will reward these companies. GDP will look strong. This is the easy part.

Medium-term (2028-2031): The demand side begins to deteriorate. Consumer-facing companies report declining same-store sales, reduced average transaction sizes, and customer acquisition costs that rise as the addressable market shrinks. Revenue growth stalls even as productivity continues to improve. The Ghost GDP gap becomes visible in earnings calls.

Long-term (2031+): The contradiction resolves — either through massive fiscal redistribution (UBI, expanded safety nets, public employment), a demand crisis that forces corporate retrenchment, or some combination of both. The investment landscape depends entirely on the policy response, which remains deeply uncertain.

The sectors most exposed to Ghost GDP risk are those that depend on broad-based consumer spending: retail, restaurants, housing, automotive, consumer finance, and healthcare. The sectors most insulated are those serving the capital-owning class or providing essential services with inelastic demand.

The central irony: the companies that create the most Ghost GDP through aggressive automation may ultimately destroy the consumer markets that sustain their revenues. This is the macro version of the tragedy of the commons — individually rational, collectively catastrophic.

Key Takeaways

• Ghost GDP is economic output that appears in national accounts but never circulates through the human economy. It is the gap between production and prosperity — and AI is widening it faster than any technology in history.

• The Engels Pause provides a direct historical parallel. During Britain's Industrial Revolution, GDP doubled while wages stagnated for 50 years. AI threatens a similar divergence, compressed into a shorter timeframe and affecting a broader range of occupations.

• GDP can grow 5% while unemployment doubles and median income falls. This is not a paradox — it is simple arithmetic when production is decoupled from human labor. Traditional economic indicators will systematically overstate the health of the economy.

• Policymakers will be flying blind. The Fed, the BLS, and Congress rely on indicators designed for an economy where production and consumption are coupled. Those indicators will mislead during the critical transition period, delaying necessary policy responses.

• The consumption paradox is the terminal risk. AI can produce without limit, but it cannot consume. An economy that produces everything but has no one who can afford to buy it is not growing — it is hollowing out, regardless of what GDP says.

• New metrics are urgently needed. Median income growth, labor share of income, effective consumer base, and velocity-adjusted output would provide a more accurate picture of economic welfare. Building political support for these metrics is as important as building the metrics themselves.

• For investors, the timeline matters. Short-term, AI productivity plays will outperform. Medium-term, the demand erosion will become visible in earnings. Long-term, the investment landscape depends on the policy response to a problem that most policymakers do not yet recognize as real.