Citrini Research's 2028 scenario contains a single sentence that is the entire fiscal story: government "must transfer more money to households at precisely the moment it is collecting less money from them in taxes." Federal receipts are projected 12% below CBO baseline. Payroll taxes are falling. Unemployment is heading to 10.2% by mid-2028. The math does not break first. Politics does.
This is the structural problem the AI displacement spiral creates for every advanced democracy, and it has no clean technical solution.
The arithmetic of a labor-share collapse
The federal revenue base is a labor-tax structure with capital taxation bolted on as an afterthought. Payroll taxes (Social Security, Medicare) are roughly a third of federal receipts. Personal income tax — also dominantly a tax on wage income — is another half. Corporate income tax, capital gains, and estate taxes together fund the remainder.
When labor's share of GDP falls from 64% in 1974 to a projected 46% by 2028, the tax base falls with it. Not in nominal terms — output is still rising, because hyperscalers spent up to $660 billion on AI capex in 2026 alone. In labor-base terms. The wages that fund payroll tax are simply not being paid in the same proportion.
The expenditure side moves the other direction at the same time. Unemployment insurance kicks in. SNAP enrollment rises. ACA subsidies expand because employer-sponsored coverage falls with employment. Medicaid grows. Citrini's scenario has jobless claims at 487,000 in Q3 2027, the highest since the pandemic.
This is not the 2008 fiscal stress. In 2008, the cyclical wave eventually reversed and tax receipts recovered. In a labor-share collapse, the wave does not reverse, because the underlying ratio is structural. The economy keeps producing output. The output simply does not pass through household balance sheets on the way to the IRS.
Why a compute tax is the obvious answer and the politically hardest one
Citrini sketches two policy responses: a "Transition Economy Act" and a "Shared AI Prosperity Act" funding dividends from an AI compute tax. The economic logic is straightforward. If labor taxation is the historical mechanism for funding the social contract, and labor is being substituted by compute, then compute should bear the tax that labor used to bear. Otherwise the social contract loses its funding source while the compute owners' share of national income compounds without limit.
Politically, this is harder than any tax fight in modern American history. The reasons stack.
The compute owners are concentrated in five firms with the highest political-spend per dollar of revenue in the economy. They have a national-security framing — "we have to win the AI race against China" — that aligns commercially and rhetorically. The chip stack is mobile in a way that factories were not; a serious compute tax in the United States is met with credible threats to relocate inference capacity to jurisdictions that will not impose one.
And the constituency for the tax is, by definition, the displaced. The displaced are not yet organized, do not yet identify as a class, and have spent decades being told their displacement is their personal fault.
The choice that gets made by default
Every advanced democracy will face this revenue paradox simultaneously, because the labor-share trajectory is global. The choices are narrow.
Option one: raise the labor tax on the workers who remain. This breaks the political coalition of the employed middle class faster than the displacement does, because the tax is concrete and visible while the displacement is statistically invisible to those still working.
Option two: borrow. This works until it does not. The bond market tolerated 100% debt-to-GDP in normal times. It will not tolerate the trajectory implied by a permanent labor-share decline funding permanent transfer payments.
Option three: cut transfers. This is what happens politically when the first two options fail and there is no political coalition strong enough to tax compute. The displaced lose their unemployment, their healthcare subsidies, and their food assistance, in that order, while the labor share keeps falling and the compute owners' wealth compounds.
Option four: tax compute, redistribute the proceeds, and accept the political fight. Every advanced democracy that survives this transition will, eventually, end up here. The question is whether they get here before or after the institutional damage of getting here through options one through three.
Which of those four paths is your political system actually capable of choosing in the next four years?