AI could shave $2.2 trillion from the deficit, but a 5-year recession could bring the debt roaring back.
AI could shave $2.2 trillion off the US deficit by 2036. But according to a new working paper from economists at Brookings and the Federal Reserve, more than half of those savings could disappear — canceling out the very disruptions AI would create.
In May, the US national debt crossed the $39 trillion mark. The difference between the money the US government spends and what it receives has become the power of financial hawks of all political stripes. Without significant reform measures from Congress, the growing deficit threatens to deplete the trust funds that fund Social Security by 2032, and Medicare one year later.
Budget experts say fixing the debt will require tax increases, entitlement cuts, or possibly a combination of the two. In the absence of that political will, AI has been floated as a financial escape. A new paper suggests that the scape hatch is smaller than advertised.
If AI leads to significant increases in productivity, higher output per worker in each economy could help increase government spending and stabilize budgets, according to a working paper published Wednesday by Brookings and Fed economists. On the revenue side, AI-driven productivity gains could mean the government could collect more from the wider economy without raising tax rates. On the spending side, AI can also help erase inefficiencies, especially in healthcare systems, where administrative costs alone account for a quarter of all costs.
Overall, the increase in AI production could reduce the country’s budget deficit from the 6% of GDP currently sitting on the floor to 2%, which is equivalent to $2.2 trillion removed from the American bill by 2036, the authors of the paper wrote. But that number comes with a quick caveat that the paper’s authors buried in their conclusion: the same AI boom could reverse more than half of those savings with a combined five negative effects.
Technology has brought such miracles before. In the 1990s, the Internet-driven stock market and economic activity growth led to a 2.2% increase in annual tax revenue as a percentage of GDP, according to a previous Brookings study. The euphoria of this decade led to a nearly 60% reduction in the deficit between 1992 and 2002.
But although the 90s started with a strong market and economic boom, it did not end that way. The gains of the dot-com-era eroded within a decade. Brookings economists warn that the AI boom could erode very quickly – and point to five ways it’s happening.
- Longer lives, higher costs
One of the most life-changing impacts of AI may be in the very definition of that word. By improving medical diagnosis, treatment procedures, and healthcare efficiency, AI can significantly reduce mortality rates. Some clinical studies tracking the impact of AI-enabled early warning systems have led to significant reductions in in-hospital patient mortality. One AI algorithm, trained to identify patients at risk of sepsis, was associated with a 17% reduction in mortality.
It is a clear public interest that cannot be argued against. But the budget lens tends to look at life—and how long it lasts—differently. Longer lives also mean more years for Americans to receive benefits from programs like Social Security and Medicare, Brookings researchers note. A decrease in mortality will result in a larger number of people of retirement age qualifying for these benefits, leading to higher spending.
The Brookings paper estimates that the most disturbing scenario could see more than 3 million people of retirement age added to the population by 2036. AI could pave the way for a healthier and longer-lived population, but that could be too expensive for the federal government to maintain.
- Tax base shifts
Highly integrated AI could cause major changes in the way the government makes its money. In the 1990s, the capital gains tax was the largest source of increased government revenue, according to an earlier Brookings study. That fits the budget because in the US wages are generally taxed more than capital gains or corporate taxes.
So far this fiscal year, individual income taxes account for 52% of all federal revenue, compared to about 6% for corporate taxes, according to the Treasury Department. Receipts from capital gains taxes are often even smaller, as many wealth-building assets are unrealized. The 2024 IRS study found that the effective tax rate on capital gains remains at about 5%.
If more of the state’s revenue is earned as profits, rents, or returns on ownership than payments, as is likely in the case of AI productivity gains, the tax rate could decrease even if the amount of revenue increases, Brookings’ authors warn. Improved productivity does not automatically translate into greater government revenues if the benefits accrue primarily to property owners rather than workers.
The result may be less tax take than policymakers would expect from headline GDP numbers alone.
- Weak staff
One of the reasons that AI-driven productivity gains can increase business profits while failing to deliver measurable gains in income tax receipts could be that fewer people earn money to pay taxes on.
Whether AI will reduce the workforce by firing workers or discourage them from participating remains an unanswered question with real implications for federal budgets. Low participation means that fewer people pay wages and income taxes, and more people rely on income support programs that the government must pay.
In disruptive AI scenarios, the Brookings authors project a 3% drop in the labor force participation rate, roughly equivalent to 6 million fewer people working by 2036 — a blow similar to that faced by the COVID-19 pandemic, but potentially permanent. This could mean millions more enrolling in programs like SNAP for food assistance or disability benefits, which weigh heavily on federal spending needs.
- High borrowing costs
By overcharging the economy, the AI buildout itself may result in higher interest rates. Large investments in chips, data centers, and supporting infrastructure may increase neutral interest rates, which in turn increase market rates and corporate debt service costs.
In a high debt environment, even a small increase in interest rates can add up to a huge financial burden. Brookings authors estimate that AI productivity could add nearly $60 billion to the cost of servicing the federal debt by 2036.
- AI arms race
Ultimately, AI may spark an expensive international arms race, which will ultimately be the cost of governments to bear. If competing countries accelerate military spending to keep up with the growing power of American companies, the US may be pressured to do the same, meaning the long-term impact of AI could pile up on spending on defense programs that are already considered expensive for the country.
Maintaining a strategic margin in the age of AI could end up adding more than $350 billion in defense surpluses to the national deficit over the next decade, according to the paper.
In total, these five reductions could recoup more than half of the fiscal gains the US could expect from the AI productivity shock — meaning the headline $2.2 trillion savings is, in reality, closer to $1 trillion or less. AI may boost the economy and slow some of the worst effects of the US deficit, but it likely won’t replace the hard work of balancing the nation’s books in the long term.
