Ad: BlueJ Better Tax Answers. -Accomplish hours of research in seconds -Instantly draft high-quality communications -Verify answers using a library of trusted tax content. Learn more

Is the AI Boom Hollowing Out Corporate Tax Receipts?

In September 2026, the Congressional Budget Office reported a 25% year-over-year decline in corporate tax receipts, based on the first eleven months of the 2026 federal fiscal year. The absolute eleven-month decline was $96 billion—from $390 billion in FY 2025 to $294 billion in FY 2026—and comes on top of a 15% annual decline from FY 2024 to FY 2025. For this year, however, overall federal receipts are up by 3%, driven principally by individual and payroll taxes. These data show that, over the last two fiscal years, the composition of federal receipts has shifted sharply away from corporate income taxes.

The cause, from CBO, is not terribly surprising. The One Big Beautiful Bill Act “allows corporations to take larger deductions for certain investments, thereby reducing some payments and offsetting the increases in those receipts that otherwise
would have been expected, given the rise in corporate income.” Even less surprising is the industry making conspicuous use of these investment-related deductions, according to Politico: hyperscalers building out AI data centers. More coverage and commentary, below the fold.

Timing is everything. Just as the OBBBA restored and made permanent 100% bonus depreciation, technology companies ramped up spending on chips, servers, racks, and other physical infrastructure to underwrite AI development. Now, midterm elections are forcing a reckoning with the “populist backlash“—justified or not—against data center development that has built over several years. The practical effects are concrete: States that previously embraced data centers are seeking to claw back benefits, rewriting their tax incentives, and worrying about their income tax bases. The “toxic politics” of data centers are spilling into federal races.

This background tees up the headline for Brian Faler’s analysis in Politico: Corporate Tax Payments Plunge as AI Feasts on New Incentives. As Faler reports:

Goldman Sachs figures AI expenditures this year will approach $600 billion in the U.S. and $1 trillion worldwide. . . .

Companies in the thick of it have been reporting big drops in their tax bills. In July, Microsoft told investors that its current tax bill amounted to $2.5 billion, down from $14.1 billion the previous year, even as its income soared.

The conventional complaint is that the OBBBA’s incentives spend a lot to buy relatively little. To the extent that technology companies would make these investments anyway, the tax benefits’ incentive effects may prove limited. As Andrew Leahey argues, “tax policies may shift the timing, scale, and location of investments, but it is much harder to argue that they created the underlying investment imperative.” That said, timing, scale, and location are crucial features in AI development—and factors that may have broader implications in the emerging geopolitical struggle over the burgeoning technology. The task, as usual, is to compare clear front-end costs—the CBO’s September numbers on corporate tax receipts—with inchoate back-end public benefits, either through increased tax revenue or in-kind social gains.

And this timing issue, of course, is the ballgame for expensing. Much of data centers’ federal benefit accelerates tax benefits that otherwise would have accrued in subsequent years. Ordinarily, that benefit unwinds over time, as business activities produce income that’s unsheltered by already-claimed deductions. But if technology companies continue their outsized investments in compute, the reversal of this year’s timing benefit may be obscured by a new round of deductions generated next year. This cycle increases the pressure on businesses’ crossover point, where taxable income becomes positive—and encourages tax gaming or political deals to mitigate these future tax liabilities.

In this story, the AI-specific wrinkle is that much of the investment in compute really is short-lived. Microsoft reports useful lives of two to six years for servers and network equipment; Alphabet generally uses six years. Amazon recently reduced its useful life estimates from six years to five, expressly citing the burgeoning pace of AI development. Servers and chips have short economic lives anyway—maybe getting shorter—and their replacement incurs real (and really enormous) cash costs.

The tax system’s depreciation schedules historically have been taxpayer-favorable, accelerating benefits across broad swaths of property. For today’s chips and servers, five-year MACRS may approximate economic depreciation more closely than it does for many longer-lived assets. Full expensing still confers a present-value benefit and heightens the associated revenue loss, but, for investment in compute, the economic story is less distorted than for hard assets with longer functional lives. On this dimension, the subsidy to AI hardware is less extraordinary than the enormous first-year deduction makes it appear.

So is the AI boom hollowing out corporate tax receipts? In the short run, plainly. As a permanent matter, it’s harder to say. The investment cycle in AI infrastructure is at an inflection point. The harder questions are the texture in how expensing breaks with the real economics of AI investment—and whether the ultimate public benefits of this investment are worth today’s tax costs.

Related TaxProf Blog coverage:


About the Author

Ad: BlueJ Better Tax Answers. Blue J's generative AI tax research solution is transforming how tax experts work. Learn more.
Information and rates on advertising on TaxProf Blog

Discover more from TaxProf Blog

Subscribe now to keep reading and get access to the full archive.

Continue reading