A growth promise meets expensive debt
Outside the recovery from the COVID-19 pandemic, the US economy has grown by 3% a year only twice this century. Yet Trump has said the economy is growing faster than ever and that growth will eventually help the government deal with its debt.
Bond investors are not buying that optimism. Last week, the yield on 10-year US Treasury bonds rose to its highest level in nearly 25 years, more than one percentage point above where it stood when Trump began the unsuccessful war against Iran.
War-driven inflation has pushed yields higher and prompted the Federal Reserve to raise short-term rates. But the government’s widening fiscal gap is also weighing on the bond market. Investors want higher returns to compensate for the growing mismatch between spending and revenue.
Borrowing could get more expensive still. Foreign central banks, once regular buyers of Treasuries, have reduced their holdings. The Treasury now relies mainly on private investors, while competing for their money with AI companies borrowing heavily to build data centers for AI agents.
The result is a feedback loop: higher yields put more pressure on the budget. Interest payments on federal debt already equal 3.3% of GDP, compared with an average of 2.1% over the previous 50 years.
The growth rates the budget would need
Trump’s budget policies make the math harder. The One Big Beautiful Bill Act is estimated to add $4.7 trillion to federal debt by 2035. The deficit is already 6% of GDP, twice the level Treasury Secretary Bessent had previously promised. The Congressional Budget Office projects it will approach 7% by 2033.
That projection predates Trump’s promise to pay $5,000 to every adult American if Republicans retain control of Congress after the midterm elections.
The Committee for a Responsible Federal Budget (CRFB) tested what it would take to bring the deficit down to 3% of GDP by 2036. If a future Congress makes Trump’s temporary 2025 tax cuts permanent, and the government does not replace revenue lost after the Supreme Court overturned tariffs imposed on national-security grounds, the economy would need to grow by an average of 4.4% a year for the next decade. A balanced budget by 2036 would require annual growth of 7.2%.
AI could, in theory, deliver extraordinary growth. Some economists have described scenarios in which it raises growth to 15% a year by taking on a substantial share of the intellectual work people now do. But those scenarios look unlikely.
CRFB estimates that stabilizing federal debt would require total factor productivity—the efficiency with which resources are turned into goods and services—to grow by an average of 2.5% a year over the next decade. The United States has reached that pace only once since 1959, despite major productivity gains from electrification, the interstate highway system, telecommunications, and the first wave of information technology and automation.
More output does not guarantee more revenue
Even a major AI-driven expansion may do less for public finances than the headline growth rate suggests. More of the gains could go to capital rather than labor, and the tax rate on capital is about half the rate on labor. Supporting workers left behind by that shift could also require substantial government spending.
AI investment itself adds another complication. Debt-funded spending by AI companies pushes interest rates up, making it harder for the government to borrow. And the industry’s revenue expectations are unusually demanding.
Stanford economist Hanno Lustig estimates that to earn a return on major AI investments, data-center owners would need revenue to grow by an average of 45% a year for the next seven years. Those investments are estimated to reach $1.43 trillion this year. By 2032, the required revenue would amount to roughly 9.2% of GDP.
Jared Bernstein, former chair of Joe Biden’s Council of Economic Advisers, and Stanford economist Ryan Cummings reach a similar conclusion. They estimate that six companies—Google, Meta, Microsoft, Oracle, SpaceX, and Amazon—would need to generate an additional $13.1 trillion to $18.7 trillion in revenue over the next 10 years to cover their AI investments. That is about as much as they earned over the previous 10 years.
I think the central question is not whether AI can make the economy grow faster. It is whether the resulting gains will reach the federal budget in time, and at a scale that can offset both the government’s borrowing and the industry’s own financing needs. If the revenue forecasts fall short, the same AI buildout being offered as an answer to the debt problem could make that problem harder to manage.
Daily AI news
Every day we pick what actually matters in AI and explain it plainly — no hype, no filler. Subscribe if you want to follow where the industry is going.
Only what matters — every day
Follow on X