The Race — Building Democracy in the Age of AI. People race toward a shared finish line.
The Race
← Trends

Supporting case

A Cleveland Fed paper tests whether taxing AI capital can finance public transfers

What happened

Published September 4 and André Victor Doherty Luduvice and Roberto Pinheiro’s working paper models AI as a distinct type of productive capital. Its fiscal experiment taxes AI capital income net of depreciation—the return after allowing for assets wearing out—and distributes additional receipts as equal per-person transfers. It is an economic simulation, not a proposed or enacted federal tax. The full paper ↗ permits a substantive assessment beyond the abstract.

The model finds limited fiscal room while AI is a small part of the capital stock. Cheaper AI expands the base; pairing that price decline with the modeled revenue-maximizing tax improves low-skill-worker and aggregate welfare. A separate roughly $12,000 annual-transfer experiment pairs an 84% AI-capital-income tax with a 28% consumption tax. It benefits low-skill workers but leaves aggregate welfare roughly neutral; those are parameters of the simulation, not recommended real-world tax rates. The results depend on calibrated assumptions about worker skills, ownership and substitution between labor and capital; they do not identify real tax receipts or a ready-to-administer AI tax base. The authors’ very high modeled tax rates are counterfactual parameters, not a policy recommendation validated by observation.

For The Race, the contribution is locating taxable surplus and examining how large a social dividend it could actually support. That is distinct from taxing each displaced job or assuming productivity pays for services automatically.

Dates and implementation stage

Working paper published September 4, 2026

Supporting sources

Related trends