Roughly half of all consumer spending in the United States comes from the top 10% of earners. That is the single statistic that makes the AI-displacement story dangerous to the broader economy, and it is the one most cheerful AI productivity decks leave out.
Forty years of widening inequality changed the postwar shape. By 2024, the share of personal consumption expenditure driven by the top decile had crept past 50% and kept climbing. When the marginal dollar of demand depends on a narrow band of high earners, anything that hits that band is no longer a slice of GDP — it is the load-bearing wall.
What 2% does to 50%
Citrini Research's 2028 scenario builds the arithmetic explicitly. A 2% employment loss concentrated in the top 10% — managers, lawyers, software engineers, consultants, financial professionals — translates to a 3–4% decline in discretionary spending. That elasticity is not a guess. It is what you get when the displaced were saving a meaningful fraction of their income, when their consumption skews more discretionary than the median household's, and when the shock hits the demographic that buys the second car, the renovation, the vacation, the private school.
The multiplier is brutal because high earners were the cushion the rest of the economy was leaning on. Restaurants in midtown Manhattan, contractors in Marin County, dentists in Austin — all downstream of that decile's discretionary income. When the senior product manager who used to spend $400 a week eating out is laid off, the line cook who served her loses hours, the line cook stops getting his car serviced, and the mechanic stops eating out himself. Economists call it the marginal propensity to consume. The neighborhood calls it the for-lease sign in the window.
What makes this cycle different from 2008 is that the trigger is structural rather than cyclical. In a normal recession, the laid-off manager is rehired into something similar. In this scenario, the role itself does not return — it is now done by a software process that costs a fraction of the salary. The displacement is permanent at the role level, and the worker has to retrain into something the same model is also coming for.
The shape of an early signal
You can already see the leading edge. White-collar job openings down 15% year over year. New computer science graduates at 7.0% unemployment, worse than the rate for high school graduates. Entry-level postings down roughly 35% since January 2023. None of these by themselves cause a 3–4% spending decline. All of them together describe a labor market where the historic on-ramp into top-decile earning has narrowed sharply, even before the displacement at the top hits.
ServiceNow's projected ARR growth slowing from 23% to 14% reads the same signal through enterprise software: companies are buying less software because they are running fewer of the workflows that used to consume the licenses. Federal payroll-tax receipts running 12% below the Congressional Budget Office baseline reads it through government revenue. The same story from different angles.
Home prices in San Francisco, Seattle, and Austin sliding 8–11% year over year in the same projection are the asset-price echo of the wage-side compression. These are not subprime markets. They are the markets where the median buyer used to be a senior engineer at a tech company. When that buyer is uncertain about their job for the next eighteen months, the bid disappears, and price discovery goes downward.
What this means for anyone budgeting toward 2028
If your business concentrates on the top 10%, the question is not whether your sales correlate to that decile's employment. It is whether your model treats that correlation as cyclical or structural. If your investor deck assumes a 2026 hiccup followed by a 2028 recovery, and the displacement turns out to be permanent at the role level, the deck is wrong about the most important variable.
The polite framing is that AI redistributes work. The arithmetic is that AI redistributes income upward fast enough to break the demand half of its own thesis. Productivity gains do not feed themselves into consumption automatically; somebody has to spend the savings. When the savings concentrate in compute owners and the displaced stop spending, the gains are real and the demand is gone. Which 10% does the next decade's discretionary spending come from when the current one is no longer employed?