AI has a $2.5 trillion problem and replacing 250 million jobs is the only way to solve it

TEXT START: When you buy through links on our articles, Future and its syndication partners may earn a commission.

The Dissection

This is a solvency alarm disguised as AI economics. It argues that roughly $2.5 trillion in annual revenue is needed to justify the buildout, while current AI revenue sits at $150–$250 billion. It then links that shortfall to mass labor substitution and exposes the infrastructure mismatch: chips become obsolete in about five years, while data centers last roughly twenty, and model prices are collapsing faster than physical costs.

The article’s real function is to show that AI capital requires a conversion of human labor costs into machine-captured revenue. Its own figures also reveal the danger to infrastructure providers: they may finance a permanent physical footprint for an industry whose software economics are rapidly deflating.

The Core Fallacy

The arithmetic is internally broken. At $100,000 per job equivalent, $2.5 trillion corresponds to 25 million jobs—not 250 million. Replacing 250 million jobs at that rate would represent $25 trillion in annual labor value. Even the stated $2.25–$2.35 trillion revenue gap corresponds to roughly 22.5–23.5 million jobs.

The article’s comparison with 159 million US nonfarm jobs therefore magnifies its own tenfold error. The headline’s claim that replacing 250 million jobs is “the only way” is also unproven. Revenue is not identical to labor cost, and customer productivity gains do not automatically become AI-provider revenue.

The text also mixes financial categories: $1.5 trillion of cumulative hardware spending, $1 trillion of additional annual investment, and $2.5 trillion of required annual revenue are treated as though they were interchangeable. They are not.

Hidden Assumptions

• Every $100,000 of labor cost can be converted into equivalent AI-provider revenue.
• Productivity gains accrue primarily to AI companies rather than customers, competitors, or consumers through lower prices.
• Job replacement is binary rather than task-based, gradual, or complementary.
• The US labor pool is the relevant ceiling despite the article’s own global-scale investment claims.
• Hardware depreciation, utilization, financing, residual value, and replacement cycles are captured by the simplified “50-Cent Rule.”
• Falling model prices will not generate enough volume or new demand to offset margin collapse.
• The current revenue estimates and $2.5 trillion target are sound despite the article not deriving them in detail.

Social Function

This is a partial truth serving transition management and elite self-exoneration. It correctly identifies a capital-recovery problem and a likely stranded-asset risk. But it presents mass displacement as an accounting necessity. That framing turns a structural transfer of power into “the math,” allowing owners to describe the destruction of labor demand as an unavoidable business requirement.

It also functions as ideological anesthesia: if AI replaces enough workers, the investment supposedly works. The consumption system that those workers supported is left outside the spreadsheet.

The Verdict

The article’s number is wrong, but its structural warning is real. Under the Discontinuity Thesis, AI becomes economically dominant by severing the labor-to-wage circuit. The same substitution that creates returns for AI owners destroys the mass purchasing power required by post-WWII capitalism.

The infrastructure providers face a separate carcass-management problem: long-lived physical assets may be trapped beneath rapidly commoditizing intelligence. The spreadsheet is defective. The corpse it is measuring is not.