AI may not deliver the full deficit relief economists project

Artificial intelligence could help shrink the U.S. budget deficit by raising productivity, but a new working paper says the savings may be far smaller than headline estimates suggest. Economists at the Brookings Institution and the Federal Reserve estimate an AI-driven boom could reduce deficits by as much as $2.2 trillion by 2036, while related costs could absorb more than half of that gain.

The warning comes as Washington faces rising pressure from debt and entitlement spending. Fortune reported that the U.S. national debt crossed $39 trillion in May, while the Peter G. Peterson Foundation has said Social Security’s trust fund is projected to be depleted in 2032 and the American Hospital Association has cited Medicare trustees projecting the hospital insurance trust fund could become insolvent in 2033.

The Brookings-Federal Reserve paper says stronger productivity could lift output per worker, expand the economy and raise tax receipts without higher tax rates. The authors also say AI could reduce inefficiencies in federal spending, including in health programs, where administrative costs account for about one-quarter of expenses.

Under a strong productivity scenario, the paper says the annual deficit could fall from about 6% of gross domestic product to as low as 2% by 2036. That would amount to roughly $2.2 trillion in deficit reduction, according to the authors.

Why the savings could fade

The same paper identifies five channels that could push costs back up. One is longevity. The authors say AI improvements in diagnosis, treatment and health care efficiency could reduce mortality, a social benefit that would also increase the number of older Americans receiving Social Security and Medicare for longer periods.

In a high-disruption scenario, the paper estimates the retirement-age population could be 3 million larger in 2036. That would raise federal benefit spending, even if AI also improves health outcomes.

A second risk is a shift in the tax base. The Treasury Department says individual income taxes account for 52% of federal revenue so far this fiscal year, compared with about 6% from corporate taxes. A 2024 IRS study found the effective tax rate on capital gains was about 5%.

The Brookings-Federal Reserve authors say government revenue may rise less than expected if AI’s gains flow more to profits, rents or ownership returns than to wages. In that case, a larger economy would not produce the same tax yield that policymakers might expect from payroll and income growth.

Labor-force effects are another concern. In disruptive AI scenarios, the paper projects a 3% drop in labor force participation by 2036, equal to about 6 million fewer workers. The authors say that would reduce payroll and income tax receipts while increasing demand for programs such as SNAP and disability benefits.

Higher interest costs could also cut into the fiscal benefit. The paper says heavy investment in chips, data centers and related infrastructure could push up the neutral interest rate, raising the government’s cost to service debt. The authors estimate AI-related productivity could add about $60 billion to federal debt-service costs by 2036.

The final risk is defense spending. The paper says AI could encourage rival countries to increase military outlays, putting pressure on the U.S. to respond. The authors estimate that maintaining a strategic edge could add more than $350 billion in cumulative defense spending over the next decade.

Brookings researchers have pointed to the 1990s as a reminder that technology booms can help the budget. Earlier Brookings research found the internet-era expansion raised annual tax revenue by 2.2 percentage points of GDP and helped cut the deficit by about 60% between 1992 and 2002. The new paper argues AI may still improve the fiscal outlook, but not enough to replace tax and spending decisions by Congress.

This story draws on original reporting from Fortune.