Citrini Research's 2028 Global Intelligence Crisis names a phenomenon orthodox macro cannot price: output that shows up in national accounts but never circulates because machines do not consume. Velocity of money flatlines. Nominal GDP looks fine. Real demand collapses underneath it.
This is the part of the AI labor transition no quarterly print captures, because every line item looks healthy.
What "ghost GDP" actually means
National accounts measure production, not circulation. When a software company replaces a 200-person customer-service team with AI at the same throughput, the value-add gets booked. The hyperscaler's compute revenue spikes. The software company's margin expands. Reported GDP is unchanged or higher.
What disappears is the next-order spending. Two hundred salaries no longer flow into rent, groceries, daycare, vacations, starter-home mortgages. The compute provider that captured the value-add does not redeploy that revenue into the same channels. It deploys it into more GPUs, datacenter buildouts, equity buybacks.
The output exists. It does not circulate. Hence: ghost.
Velocity of money is the variable orthodox models trust to mean-revert. The displacement spiral breaks that assumption. When wages convert to capex, the dollar moves once, into a Nvidia order, and sits as depreciable equipment for five years.
Big-four hyperscaler AI capex for 2026 is projected at up to $660 billion. The capital is real, the production is real, and most of it never reaches a household.
The signatures already showing up
Federal receipts running 12% below CBO baseline, with payroll taxes falling — that is the fiscal shadow of vanished wage income. The capex shows up as corporate revenue and capital deduction. The wages it displaced do not show up as payroll tax at all.
Home prices in the metros most exposed to white-collar displacement: San Francisco down 11% year-over-year, Seattle down 9%, Austin down 8%. These markets are where the ghost is loudest, because the salaries that supported the price stack are being eliminated. The junior software developer cohort, down roughly 20% since late 2022, is the same cohort that was supposed to take out next year's mortgages.
White-collar job openings down 15% year-over-year. ServiceNow ARR growth decelerating from 23% to 14%. Mastercard revenue growth dropping to 3.4% as consumer agents route around card rails. Each is a micro-story. Together they are the signature of money no longer circulating at the velocity models assume.
Acemoglu's research already showed automation explains 50 to 70% of US wage inequality from 1980 to 2016. Ghost GDP is what that mechanism looks like at the macro print once displacement runs faster than reskilling.
Why this breaks the political consensus
The orthodox playbook for a slowdown is fiscal stimulus plus rate cuts. Ghost GDP makes both weaker than usual.
Fiscal stimulus assumes the government can collect from the booming sector and transfer to the displaced one. But the booming sector — hyperscaler compute and a handful of frontier AI labs — has the capital to lobby, the legal sophistication to minimize taxable income, and a national-security framing Congress is reluctant to touch. Citrini names the paradox: the government "must transfer more money to households at precisely the moment it is collecting less money from them in taxes."
Rate cuts assume cheaper credit revives demand. In the displacement spiral, cheaper credit funds more AI capex, which compounds the displacement. The brake hits the accelerator.
Labor share of GDP has fallen from 64% in 1974 to 56% in 2024, with Citrini projecting 46% by 2028. That is the same number from a different angle: a smaller share of national output flowing to households, a larger share to compute owners.
Reported GDP can hold up while half the country quietly stops being able to afford anything. The headline numbers will keep saying the economy is fine. The streets will say something different.
What gets built or doesn't
Two policy responses are on the table. A Transition Economy Act that retrains displaced white-collar workers. A Shared AI Prosperity Act that taxes compute and routes the receipts to dividends, re-coupling the AI productivity boom to household income.
Neither is technically exotic. Both are the digital-era version of the labor and welfare bargains that put the post-1945 middle class together. The question is whether they pass in 2026 or after the November 2027 perception flip, when the S&P is down 38% and unemployment is closing on 10%.
There is a third option. Do nothing. Let nominal GDP tick along, let labor share keep falling, let velocity grind toward a number macro forecasters have no model for. Watch the print stay green while the economy underneath stops circulating. Whose problem is that, when the line still goes up?