A GPU cluster in a data centre produces the analytical output of 10,000 knowledge workers. It does not eat lunch. It does not commute. It does not pay rent or buy its kids shoes.
GDP measures output. When the GPU cluster replaces the workers, output is maintained. GDP holds. Productivity soars. The quarterly numbers look excellent.
But the income that those 10,000 people spent — at restaurants, on rent, on shoes — that income is gone. The output exists. The demand it used to fund does not.
We call this Ghost GDP: economic output that appears in the national accounts but never circulates through the real economy as consumer spending.
This is not a thought experiment. It is an observable, measurable phenomenon already visible in our data. This article explains the mechanism, shows what our numbers reveal, and argues that Ghost GDP is the single most important macroeconomic concept that mainstream economists are not yet tracking.
The Arithmetic Nobody Wants to Do
Start with the accounting identity, because accounting identities do not lie.
Consumer spending is roughly 68% of US GDP. In the UK it is 63%. In the Eurozone it averages 54%. In Japan it is 55%. Across the developed world, consumers are the economy. Everything else — government spending, business investment, net exports — is secondary.
Consumer spending is a function of consumer income. Consumer income is overwhelmingly derived from wages. Yes, there is investment income, government transfers, and gig economy earnings — but for the median household in every developed country, the wage is the engine.
Our dataset tracks 2,860 occupations across 39 countries. Roughly 41% are rated HIGH or CRITICAL risk. These occupations are disproportionately in the upper half of the income distribution. Professional services. Analytical roles. Mid-management. Administrative functions. Financial operations.
The top 20% of earners in the US account for roughly 40% of all consumer spending. These people are employed in exactly the occupations our model flags as elevated risk.
Do the arithmetic. If AI displaces 15% of HIGH-risk roles over the next three years — a rate consistent with what our employment change data shows for the most exposed occupations — the direct impact is not 15% of their total spending. It is the marginal spending they cut first.
Nobody stops paying their mortgage immediately. Nobody pulls their kids out of school. What they stop doing is: the dinners out. The kitchen renovation. The weekend trip. The second car lease. The boutique gym membership. The subscription boxes. The impulse purchases.
This marginal spending is the lifeblood of the service economy. It funds restaurants, retail, travel, entertainment, personal services — exactly the sectors that employ the workers who aren't being displaced by AI. When the displaced professional class cuts its discretionary spending, the ripple effect hits workers who have never heard of GPT.
This is the Ghost GDP mechanism: AI increases output while decreasing the income that fuels demand. The numerator grows. The denominator shrinks. The ratio looks like productivity. The reality is a slow-motion demand crisis.
Why GDP Is the Wrong Metric
GDP was designed in 1937 by Simon Kuznets for the US Congress. Kuznets himself warned that "the welfare of a nation can scarcely be inferred from a measurement of national income." He was ignored then. He is being ignored now.
GDP measures gross output. It does not measure:
- Who gets paid. If a company replaces 100 workers with an AI system and produces the same output, GDP is unchanged. The company's profits are up. The workers' income is zero. GDP does not register this as a problem.
- How income circulates. A dollar earned by a middle-class worker is spent 4-5 times in the local economy (the fiscal multiplier). A dollar of corporate profit may be retained, invested in more AI infrastructure, distributed to shareholders, or parked overseas. The multiplier is lower. GDP counts both dollars the same.
- The distribution of output. GDP per capita in the US is $83,000. Median household income is $80,000. The average conceals a distribution that is increasingly concentrated at the top. AI accelerates this concentration by shifting income from labour (wages) to capital (profits).
Every quarter, mainstream financial media runs the same story: "GDP grew at X%. The economy is strong." And every quarter, the story misses the structural decay happening underneath.
Our data provides the missing layer. By tracking 2,860 occupations simultaneously, we can see what GDP cannot: the occupation-level mechanics of how output growth decouples from income distribution.
The Six-Month Fuse
High-income workers have savings buffers. This is both obvious and critically important for understanding the timing of Ghost GDP's impact.
When a senior product manager earning $180,000 loses their job, they do not immediately stop spending. They have six months of savings, maybe twelve. They have a working spouse, maybe. They have unemployment insurance, a 401(k) they can draw from in extremis, a credit card with a $30,000 limit.
For three to six months, their spending barely changes. They tell themselves it is temporary. They network. They apply for jobs. They update their LinkedIn. They maintain the consumption patterns of an employed professional, because they still think of themselves as one.
This is why labour market disruptions are so dangerous: the demand destruction is lagged. By the time the spending contraction shows up in retail sales data and consumer confidence surveys, the structural damage has already been inflicted. The workers who were displaced six months ago are now drawing down savings. The workers who will be displaced next quarter are still spending normally. The aggregate data shows a consumer economy that looks fine — until it doesn't.
This is exactly what our data is designed to detect. We track wage change and employment change as separate signals because they tell different stories.
Employment change tells you the stock — who has a job today. Wage change tells you the flow — what the people who still have jobs are being paid. When both decline simultaneously for the same occupation cluster, the consumer demand impact is not hypothetical. It is imminent.
In our dataset: wage change coverage is 96% of tracked occupations. Employment change coverage is 99.5%. We see the convergence in near real-time. Most economists will see it six months later in the consumer confidence surveys. Twelve months later in the GDP revisions.
The Absorber Problem
Every automation wave produces the same reassurance: new industries will emerge, new jobs will be created, it will all work out. And every time, for 250 years, it has. Agricultural workers moved to factories. Factory workers moved to offices. Office workers moved to... where?
This is the question that the "it always works out" crowd cannot answer, because the historical analogy is broken in a specific, measurable way.
Previous automation waves displaced workers in one sector and absorbed them in another. The absorber sector was always larger, more productive, and higher-paying than the displaced sector. This is why automation was, on net, positive for workers: the destination was better than the origin.
The absorber for every wave since 1950 has been the professional services economy. Consulting. Analysis. Administration. Finance. Law. Marketing. HR. IT services. Education. Healthcare administration. Government services. The entire white-collar infrastructure that runs the modern economy.
AI is displacing the absorber sector itself.
A factory worker in 1985 could retrain as an office administrator. A coal miner in 2005 could retrain as a data entry clerk. The retraining pipeline was clear: learn computer skills, get an office job, join the professional services economy.
A displaced office administrator in 2026 can retrain as — what? The roles above them require decades of institutional trust and relationship capital. The roles below them pay less and offer no career progression. The roles sideways are being automated by the same technology that displaced them.
This is not speculation. It is the structure of the displacement pattern, visible in our monthly data. We can see which occupations are losing employment, which are gaining, and what the characteristics of each group are. The gaining occupations are overwhelmingly physical-presence, manual, or high-trust interpersonal roles. The losing occupations are overwhelmingly cognitive-routine professional roles.
The absorber class is the exposed class. That has never happened before.
What Happened to the Multiplier
Economists talk about the "fiscal multiplier" — the number of times a dollar circulates through the economy before it is saved. For consumer spending by middle-class households, the multiplier is roughly 4-5x. You spend $100 at a restaurant. The restaurant pays its staff, who pay their rent, who fund their landlord's grocery shopping, and so on.
For corporate profits, the multiplier is lower. Profits may be:
- Retained as cash on the balance sheet
- Invested in capital expenditure (including AI infrastructure)
- Distributed to shareholders as dividends or buybacks
- Moved to lower-tax jurisdictions
The multiplier for corporate profit redistribution to shareholders is roughly 2-3x, because shareholders tend to save a larger fraction of additional income than wage earners do. The multiplier for retained earnings invested in AI infrastructure is essentially zero for local economies — the money goes to cloud computing providers, chip manufacturers, and AI companies, not to the local restaurant.
When AI shifts income from wages (multiplier 4-5x) to profits (multiplier 2-3x or lower), the total demand circulating in the economy falls even if GDP is unchanged.
This is not theoretical. The historical precedent is the "productivity-pay gap" that opened in the United States around 1979 and has been widening ever since. Since 1979, US productivity has grown roughly 60%. Real median wages have grown roughly 16%. The difference — the gap between what workers produce and what workers are paid — has been captured as corporate profit and capital income.
AI is accelerating this gap at unprecedented speed. The companies deploying AI are seeing immediate productivity gains that flow directly to the bottom line. The workers displaced by AI are seeing immediate income losses that flow directly to reduced consumer spending.
Our data quantifies this at the occupation level. For each of the 2,860 occupations we track, we can see the relationship between AI exposure score, wage trajectory, and employment trajectory. The pattern is consistent: higher exposure → flatter wages → declining employment. The productivity gain exists. The pay rise does not.
The Housing Market Transmission
Ghost GDP has a specific and devastating transmission mechanism through housing markets, and it is already becoming visible in certain metropolitan areas.
The professional class — the soft middle — is the primary driver of housing demand in urban and suburban markets in every developed economy. They are the people who buy houses in their 30s, trade up in their 40s, and fund the entire mortgage-backed securities ecosystem that underpins the financial system.
When professional services wages compress, the housing market transmission works like this:
- Mortgage qualification tightens. Banks use income-to-debt ratios. Lower income means lower maximum mortgage. The marginal buyer is priced out.
- Trade-up activity stalls. Existing homeowners who planned to sell and buy something larger delay, because their income growth no longer supports the jump.
- New construction demand softens. Developers who planned subdivisions based on projected professional-class income growth find their assumptions invalid.
- Property tax revenue declines. Municipalities that fund schools, police, and infrastructure from property taxes see their revenue base erode as home values stagnate.
None of this shows up in GDP as a crisis. Home prices do not crash (because supply is also constrained). New construction does not collapse (because institutional investors fill the gap). The housing market does not break — it just quietly shifts from being an engine of middle-class wealth creation to being a mechanism of wealth concentration.
The sociologist Matthew Desmond has documented extensively how housing costs consume an increasing share of lower-income households' budgets. Ghost GDP extends this dynamic upward into the professional class: as wages compress, housing costs that were manageable at $150,000/year become burdensome at $120,000/year. The displacement does not produce homelessness. It produces financial fragility — the condition of being one missed paycheque away from crisis, a condition that was historically confined to the working class but is now migrating into the professional class.
The Corporate Earnings Paradox
Here is a question that should be keeping portfolio managers awake at night: how long can corporate earnings grow while the consumer base that funds those earnings is shrinking?
The answer, based on historical precedent, is: longer than you think, but not forever.
In 2025-2026, the story looks like this:
- Companies deploy AI → costs decrease → margins improve → earnings beat expectations
- Wall Street rewards the earnings beat → stock price rises → wealth effect makes shareholders feel rich
- Meanwhile: displaced professional workers cut spending → service economy softens → but this takes 6-12 months to appear in earnings
- Meanwhile: AI-augmented workers produce more → productivity metrics improve → GDP growth looks healthy
This is a temporal arbitrage: the cost savings from AI displacement are immediate, but the demand destruction from reduced consumer spending is lagged. Companies are front-running the benefit and deferring the cost. Every quarter that this continues, the gap between reported earnings and sustainable earnings widens.
The historical parallel is the US housing market in 2005-2006. Housing-related profits looked excellent. Mortgage origination was booming. Bank earnings were strong. The underlying asset quality was deteriorating quarter by quarter, but the deterioration was hidden by the lag between origination and default. When the lag converged, the correction was not gradual. It was catastrophic.
We are not predicting a financial crisis. We are observing a structural analogy: a lag between reported profitability and underlying demand sustainability, driven by a mechanism (AI displacement) that the standard economic indicators are not designed to detect.
What Ghost GDP Looks Like in the Numbers
Ghost GDP does not look like a recession. It looks like everything is fine — until it is not.
Our data tracks four simultaneous signals that, taken together, constitute a Ghost GDP indicator:
-
Productivity up, wages flat or falling in the same occupation. We see this in 67% of HIGH-risk occupations. The output is growing. The pay is not.
-
Vacancy counts stable, employment declining in the same occupation. This is ghost demand — the illusion of a healthy labour market created by compositional shifts in job postings.
-
Corporate earnings beating expectations while consumer confidence weakens. S&P 500 earnings have beaten consensus estimates in 78% of quarters since 2023. Consumer confidence indices have been declining for the same period.
-
GDP growth positive while real median income stagnates. US GDP grew 2.8% in 2025. Real median household income grew 0.3%.
The gap between the output metrics (GDP, productivity, earnings) and the income metrics (wages, confidence, savings) is Ghost GDP. It is growing. Every month, our data measures the occupation-level mechanics of how it grows.
International Comparisons
Ghost GDP manifests differently in different economies, because institutional structures mediate the transmission.
United States: The clearest case. Minimal institutional buffers, at-will employment, weak unions. Displacement transmits to income loss within months. Ghost GDP is most visible here because the lag is shortest.
United Kingdom: Similar to the US but with slightly stronger safety nets (NHS, more generous unemployment benefits). Ghost GDP is building but the lag is 3-6 months longer than in the US.
Germany: Strong institutional buffers (Kurzarbeit short-time work schemes, works councils, industry-level collective bargaining). Ghost GDP is accumulating but the employment effects are delayed by institutional protections. This does not prevent Ghost GDP — it delays its appearance in the statistics.
Japan: Unique case. Lifetime employment norms mean that even when AI makes roles redundant, workers are often reassigned rather than dismissed. This prevents Ghost GDP from appearing in employment statistics but accelerates its appearance in wage statistics — companies keep workers but reduce their pay or hours.
Nordic countries: Strongest institutional buffers. Active labour market policies, high union density, generous retraining programmes. Ghost GDP builds most slowly here because displaced workers are quickly re-engaged. But each quarter the new roles are slightly worse (lower pay, less stable, further from original skill set) than the roles they replaced.
In every case, Ghost GDP exists. The only variable is how quickly it becomes visible in which metric. Our model, by tracking multiple metrics simultaneously across 39 countries, sees the convergence before any single-metric analysis can.
The Policy Vacuum
Ghost GDP is a new phenomenon, and the policy tools designed to address it do not yet exist. This is not because policymakers are incompetent. It is because the standard macroeconomic framework was built for a world where output and income moved roughly together.
Monetary policy cannot fix Ghost GDP. Lowering interest rates stimulates borrowing and investment. But if the investment goes into AI infrastructure that displaces more workers, lower rates accelerate Ghost GDP rather than reversing it.
Fiscal policy partially addresses Ghost GDP. Government spending can replace lost consumer demand directly — through transfer payments, public employment, infrastructure investment. But fiscal policy operates at the aggregate level. Ghost GDP is an occupation-level phenomenon. A fiscal stimulus that creates construction jobs does not help a displaced financial analyst.
Trade policy is irrelevant. Ghost GDP is not caused by foreign competition. It is caused by domestic technology adoption. Tariffs do not help. Import restrictions do not help. This is displacement from within.
The policy tool that would address Ghost GDP most directly is one that does not exist in any major economy: a mechanism for redistributing productivity gains from AI adoption back to the workers and communities affected by displacement. Some version of this has been proposed under various names — universal basic income, AI dividend, automation tax, digital services tax — but none has been implemented at scale.
Our data provides the foundation for such a policy: occupation-level, country-level measurement of displacement, wage compression, and demand destruction. You cannot solve a problem you cannot measure. We measure it.
What Comes Next
Ghost GDP is not a crisis that will announce itself. It is a slow divergence that widens quarter by quarter, invisible in the standard indicators, visible only in granular labour market data like ours.
The traditional economic indicators will catch up eventually. Consumer confidence surveys will decline. Retail sales growth will decelerate. The savings rate will drop as displaced workers draw down reserves. Housing market activity will soften in professional-class suburbs. Municipal tax revenues will stagnate.
By the time these indicators converge, the structural damage will be 12-18 months old. The workers will have been displaced. The income will have evaporated. The multiplier effect will have propagated through local economies.
We are not forecasting a recession. We are forecasting something more insidious: a permanent widening of the gap between output and income, between GDP and wellbeing, between the economy as measured and the economy as lived.
The ghost is already in the machine. The question is how long the machine can run before it notices.
We will keep tracking. You should keep watching.
Data: BLS (OES, JOLTS, CES), Eurostat (LFS, LCI, JVS), OECD, ONS, ABS, Statistics Canada, e-Stat, Adzuna API (18 countries). Risk model updated monthly. Ghost GDP estimates are derived from occupation-level wage and employment data, not aggregate national accounts. This is quantitative analysis, not financial advice.