Why an Equity Stake Beats Token Taxes for AI Automation Policy

A proposed federal tax on AI tokens has reignited debate over how governments should capture value from automation while protecting workers. But the technical details reveal why taxing AI usage may be the wrong approach—and why public equity ownership deserves closer examination.

Rep. Greg Casar (D-Texas) introduced the AI Tax and Work Protection Act to address a genuine problem: firms that use AI to reduce their workforce shift unemployment costs onto workers and society. The bill would tax AI companies based on the value of tokens processed or related revenue, with rates tied to the unemployment rate.

The token measurement problem

The proposal treats tokens—the technical units AI systems use to process information—as a taxable commodity. But tokens aren't standardized goods sold in transparent markets. They encompass text, code, images, audio, and video, with wildly different values and applications.

The Treasury Department would need to continually value these unstable units while determining whether AI companies caused labor market contractions or whether unemployment stemmed from unrelated economic shocks. The bill includes a safety valve allowing Treasury to adjust rates during wars or pandemics, but that requires real-time causal analysis the tax code isn't designed to perform.

More fundamentally, the tax creates a mismatch: companies generate taxable transactions whether they're displacing workers, augmenting them, or using AI for tasks with no labor impact whatsoever.

Why it matters

As AI reshapes labor markets, policymakers face a choice between complex usage-based taxes that require constant valuation and enforcement, or ownership structures that automatically scale with economic gains. The approach chosen will determine whether governments can fund worker protections without creating administrative nightmares or stifling beneficial AI applications.

The equity alternative

Legal scholars Jeremy Bearer-Friend and Sarah Polcz have proposed a different model: requiring AI companies to transfer equity stakes to the government rather than paying cash taxes. Sen. Bernie Sanders (I-Vt.) recently incorporated a more aggressive 50% equity transfer into legislation creating a sovereign wealth fund.

This approach offers several advantages. If AI produces modest gains, the public captures modest value. If it generates extraordinary returns, the public participates proportionally. The government values companies once rather than continuously pricing computation units.

Equity ownership is still a proxy—share prices track profits, not pink slips. A company could augment workers profitably and hand the public a windfall despite causing no harm. But it's a better proxy than token usage because it scales with economic reallocation without requiring the Treasury to determine whether each AI transaction displaced a worker.

The equity model faces its own implementation challenges: defining covered firms, valuing private companies, executing ownership rights, and determining whether government should be an active or passive shareholder. But these are one-time design questions rather than ongoing operational burdens.

Congress doesn't need to embrace Sanders' 50% stake to adopt the underlying principle. A smaller equity assessment limited to the largest firms and managed through an independent vehicle could let the public share AI's upside without overwhelming a staff-reduced Treasury Department.

The question isn't whether to capture value from AI-driven automation, but whether taxing billions of individual transactions is more tractable than owning a piece of the companies that benefit most.

These details were first reported by Bloomberg Tax in a column by Andrew Leahey, assistant professor of law at Drexel Kline School of Law.