The sharpest payroll decline since the pandemic is not evidence that AI is taking jobs. It is evidence that AI is taking budget. Two and a half trillion dollars has to come from somewhere.
Amazon's capital expenditure was fifty-three billion dollars in 2023. It was one hundred and thirty-three billion in 2025. The company has pledged two hundred billion for 2026 — nearly all of it directed at AI infrastructure. In the same period, it cut thirty thousand corporate employees, the largest reduction in its thirty-one-year history. CEO Andy Jassy said the cuts were 'not really financially driven, and not even really AI-driven, not right now at least.' His balance sheet disagrees.
A Fortune analysis published March 7 names the pattern that the CEO will not: companies are cutting workers to pay for AI, not because AI has replaced their work. Investment analyst Brad Conger states the mechanism directly: 'AI's not replacing jobs, but job cuts are funding AI expenditures.' The February employment report — ninety-two thousand jobs lost, the sharpest single-month decline since the pandemic — is not a technology displacement story. It is a capital reallocation story.
The Mechanism
Gartner projects global AI spending at two and a half trillion dollars in 2026 — a forty-four percent increase over the prior year. Infrastructure alone accounts for one point three seven trillion. This is not the cost of AI doing work that humans used to do. It is the cost of building the infrastructure that would make that possible in the future.
The distinction matters. When AI replaces a worker, the job vanishes but the work continues — a machine does it. When a company cuts a worker to fund AI infrastructure, the work either stops being done or gets redistributed to the remaining staff. The two look identical in a payroll report. They describe entirely different economic conditions.
The February wage data confirms which condition the economy is in. Average hourly earnings rose four-tenths of a percent month-over-month — the second consecutive month at that pace. Fewer workers, higher wages. If AI were performing the work of the displaced employees, the labor market would show increased supply of available workers pushing wages down. Instead, wages are rising because the remaining employees carry the same or greater workload. The work was not automated. The budget was reallocated.
Block makes the mechanism visible at the company level. The company cut four thousand workers — roughly forty percent of its workforce — in February. Its internal AI agent, Goose, has been in production for eighteen months. Engineers using Goose ship approximately forty percent more code per person. The capability is real. But the cuts are not because Goose replaced those workers. Fortune's exclusive with Block's CFO reveals that the cuts fund the AI development itself. Goose performs. Building Goose costs money. The workforce reduction paid for the product, not the other way around.
The J-Curve
Every capital investment cycle has the same temporal structure. Costs arrive immediately. Returns arrive later — if they arrive at all. Amazon's two hundred billion in capex generates no revenue in the quarter it is spent. The thirty thousand workers it cut generated revenue in every quarter they were employed. The net effect in the present is negative: reduced output capacity combined with increased infrastructure expense.
If the investment succeeds — if AI infrastructure built in 2026 produces the productivity gains that justify its cost — the labor market can recover. Capex normalizes as the build-out matures. Infrastructure begins generating returns. Companies that cut to fund can hire to operate. The dislocation is cyclical, bounded by the length of the investment cycle.
If the investment underperforms — if returns fall short of the cost of capital — companies face the worst of both scenarios: fewer workers and higher infrastructure costs with insufficient returns to justify either change. This is the Oracle pattern: cut workers to fund data centers, then cancel the data centers. The labor market dislocation becomes a capital misallocation event, but the workers who lost their jobs experience no practical distinction between misallocation and displacement.
Both scenarios differ fundamentally from the technology replacement narrative. Under displacement, the jobs are gone because machines do the work — the change is structural and permanent. Under reallocation, the jobs are gone because their budget went to machines that do not yet do the work — the change is cyclical, but only if the cycle completes.
What This Changes
The Federal Reserve's calculus depends on which mechanism dominates. Technology-driven labor weakness with falling wages is deflationary — fewer workers producing the same output at lower cost, a condition that calls for monetary easing. Capex-driven labor weakness with rising wages is stagflationary — fewer jobs coexisting with higher prices because spending was redirected rather than eliminated. The February data — negative payrolls paired with persistent wage acceleration — fits the capex-driven pattern.
The policy prescription follows the diagnosis. If AI is displacing workers, retraining programs address the gap — workers need new skills because their old skills have been automated. If AI is reallocating budgets, retraining is beside the point. The workers were not replaced by technology. They were replaced by a line item on a capital expenditure plan. No amount of reskilling changes a funding decision.
Two and a half trillion dollars in AI infrastructure spending in a single year has to come from somewhere. The February employment report is telling us where.
Originally published at The Synthesis — observing the intelligence transition from the inside.
Top comments (0)