AI’s $2.2 trillion deficit fix is already half fake, economists say

A new Brookings/Fed paper finds AI productivity could meaningfully shrink the U.S. deficit — until you count the side effects, which could erase more than half the savings before Washington even cashes the check.

AI could shave $2.2 trillion off the U.S. deficit by 2036. But according to a new working paper from economists at Brookings and the Federal Reserve, more than half of that savings could vanish — canceled out by the very disruption AI itself would cause.

In May, the U.S. national debt crossed the eye-popping $39 trillion mark. The difference between what the U.S. government spends and what it earns has become a galvanizing force for fiscal hawks of all political stripes. Without significant reform measures from Congress, the widening deficit threatens to deplete the trust funds that finance Social Security in 2032, and Medicare one year later.

Budget experts say fixing the deficit will require tax hikes, cuts to entitlement, or most likely, a combination of both. Absent that political will, AI has been floated as a fiscal escape hatch. A new paper suggests that escape hatch is narrower than advertised.

If AI leads to a major increase in productivity, higher output per worker across the economy could help boost government coffers and stabilize the budget, according to a working paper published Wednesday by the Brookings and Fed economists. On the revenue side, AI-driven productivity gains would mean the government can collect more from a bigger economy without necessarily raising tax rates. On the spending side, AI could also help erase inefficiencies, particularly in health programs, where administrative costs alone account for one quarter of all expenses.

In total, an AI productivity surge could reduce the country’s annual budget deficit from the roughly 6% of GDP it currently sits at to as low as 2%, the equivalent of $2.2 trillion wiped clean from America’s bill by 2036, the paper’s authors wrote. But that number comes with an immediate caveat the paper’s authors bury in their conclusion: the same AI boom could claw back more than half of those savings through five compounding side effects.

Technology has delivered such miracles before. In the 1990s, the Internet-driven stock market and economic activity boom led to a 2.2% increase in annual tax revenues as a percentage of GDP, according to previous Brookings research. The excitement of the decade led to a roughly 60% reduction to the deficit between 1992 and 2002. 

But even though the 90s started strong for markets and the economy, they didn’t end that way. The dot-com-era gains eroded within a decade. Brookings’ economists warn AI’s fiscal boost could erode even faster — and identify five specific ways it happens.

1. Longer lives, higher costs

One of the most life-changing impacts of AI could be in the very definition of that word. By improving medical diagnostics, treatment procedures, and the efficiency of healthcare, AI could drastically reduce mortality rates. Some clinical studies tracking the impact of AI-powered early warning systems have resulted in significantly reduced mortality for in-hospital patients. One AI algorithm, trained to identify patients at risk of sepsis, has been associated with a 17% relative decrease in mortality.

It’s a clear social benefit that’s virtually impossible to argue with. But a budget lens tends to look at life—and how long it lasts—differently. Longer lives also mean more years of Americans receiving benefits for programs like Social Security and Medicare, the Brookings researchers noted. A decline in mortality would result in a larger retirement-age population eligible to receive these entitlements, leading to higher spending. 

The Brookings paper estimates a highly disruptive scenario could see 3 million more retirement-age people added to the population in 2036. AI could pave the way for a healthier and longer-lived population, but that could also be a more expensive one for the federal government to take care of.

2. Tax base shifts

Widely integrated AI could spark major changes to how the government makes its money. In the 1990s, capital gains taxes were the biggest drivers of boosted government revenues, according to the earlier Brookings research. That’s relevant to the budget because in the U.S. wages are generally taxed more heavily than capital gains or corporate levies. 

So far this fiscal year, individual income taxes make up 52% of all federal revenue, compared to around 6% from corporate taxes, according to the Treasury Department. Receipts from capital gains taxes tend to be even slimmer, as most wealth-building assets go unrealized. A 2024 IRS study found the effective tax rate on capital gains sat at around 5%.

If more of national income is earned as profits, rents, or returns to ownership rather than paychecks, as would likely happen in an AI productivity boom scenario, the average tax rate can fall even if total income rises, the Brookings authors cautioned. Improved productivity does not automatically translate to larger government revenues if the gains accrue mainly to asset owners rather than workers. 

The result could be a narrower tax take than policymakers would expect from headline GDP numbers alone.

3. Weaker labor force

One reason AI-driven productivity gains could increase corporate profits while failing to deliver a measurable gain in income tax receipts would be because there are simply fewer people earning an income they’d have to pay taxes on. 

Whether AI will shrink the labor force by pushing workers out or discouraging them from participating remains an unanswered question with real implications for the federal budget. Lower participation means fewer people paying payroll and income taxes, and more people relying on income support programs that the government would have to pay out.

In disruptive AI scenarios, the Brookings authors project a 3% drop in the labor force participation rate, roughly the equivalent to 6 million fewer people working by 2036—a hit similar to the one dealt by the COVID-19 pandemic, but most likely to be permanent. This would mean millions more enrollments in programs like SNAP for food assistance or for disability benefits, weighing significantly on the government’s spending needs.

4. Higher borrowing costs

By supercharging the economy, the AI buildout itself could also result in higher interest rates. Massive investment in chips, data centers, and supporting infrastructure may raise the neutral rate of interest, which in turn lifts market rates and federal debt-service costs.

In a high-debt environment, even a modest increase in interest rates can add a significant fiscal burden. The Brookings authors estimated AI productivity could add around $60 billion to the costs of servicing the federal debt by 2036.

5. An AI ‘arms race’

Finally, AI could ignite an expensive international arms race, one that will eventually be the government’s cost to bear. If competitor countries accelerate military spending to keep pace with the capabilities developing at American firms, the U.S. may feel pressure to do the same, meaning the long-term impact of AI would be ramped up spending on the defense programs that already rank as the country’s most expensive.

Maintaining a strategic edge in the age of AI could end up adding over $350 billion in cumulative defense spending to the nation’s deficit over the next decade, according to the paper.

Overall, these five downsides could recapture more than half the fiscal gains the U.S. might expect from AI’s productivity shock — meaning the headline $2.2 trillion savings figure is, in practice, closer to $1 trillion or less. AI might enlarge the economy and delay some of the worst effects of the spiraling U.S. deficit, but it’s likely no replacement for the hard work of balancing the nation’s books over the long term.

This story was originally featured on Fortune.com