Before Washington starts drafting new taxes on AI-generated wealth, it should probably fix the tax system it already has. That, in short, is the message from Martha Gimbel, executive director of the Yale Budget Lab, who made the case in a Bloomberg interview on August 12, 2026.
The argument is less about AI and more about a structural problem hiding in plain sight: labor income and capital income are not taxed the same way, and that gap is about to matter a lot more as AI shifts economic output toward the latter.
The $216 billion question
The Yale Budget Lab published a report on July 20, 2026 modeling several scenarios of AI-driven economic growth. Under a rapid AI adoption scenario, the lab projects federal tax revenues could rise by up to $216 billion by 2030.
If the gains from AI productivity were distributed more evenly between labor income and capital income, rather than flowing disproportionately toward capital, the projected revenue increase could be roughly twice as large. The reason is straightforward: wages get taxed at higher rates than investment returns, and AI, by its nature, tends to generate wealth through capital rather than paychecks.
Gimbel’s point is that this gap existed before AI became a serious policy conversation, and addressing it is overdue regardless of what happens with large language models. New AI-specific tax proposals built on top of the current structure would inherit the same distortions.
What the tax code actually misses
The Budget Lab report flags two specific features of the existing code that reduce how much revenue the government captures from capital-heavy growth. First, unrealized capital gains, meaning the increase in an asset’s value before it is sold, are not taxed as they accrue. An investor whose portfolio doubles in value owes nothing until they sell.
Second, various retirement savings vehicles receive preferential tax treatment that reduces the effective rate on capital income further. Together, they mean that AI-driven growth concentrated in asset values will generate meaningfully less federal revenue than the same growth concentrated in wages.
AI’s debt relief promise, with fine print
In a May 6, 2026 analysis, the lab examined whether AI productivity could help the United States manage its long-term debt trajectory. AI-driven productivity gains could, in theory, expand the tax base enough to slow the growth of the national debt. The lab was careful to note, though, that this optimistic scenario depends heavily on how the gains are distributed and whether the tax system is positioned to capture them. Labor market disruptions, another potential side effect of rapid AI adoption, could simultaneously reduce payroll tax receipts and increase spending on safety-net programs, partially offsetting the gains.
Why this matters beyond the budget math
The Yale Budget Lab was co-founded in 2024 by Gimbel alongside Natasha Sarin and Danny Yagan, with a focus on the fiscal and distributional effects of federal policy.
If AI wealth accumulates primarily among capital owners and the tax code is structured to let capital income grow with minimal friction, the fiscal impact of AI is not just smaller than it could be. It is also less evenly distributed than it could be.
Gimbel’s prescription—fixing the existing capital income gaps first—is the more technically defensible path. The provisions she is describing, unrealized gains treatment and retirement account preferences, have survived decades of reform attempts precisely because they have broad constituencies.