Five Ways Congress Could Tax AI — and What Each Would Mean

As artificial intelligence reshapes the economy, a bipartisan group of lawmakers is exploring whether — and how — to tax the technology. The debate spans concerns about labor displacement, revenue loss from automation, and the concentration of AI wealth among a handful of companies.

The Bipartisan Policy Center has mapped the landscape of proposed AI tax mechanisms, identifying five distinct approaches that differ sharply in what they target and what consequences they might trigger. The organization has not endorsed any option but warns that choices made now will shape American competitiveness and fiscal health for years.

Why it matters

AI's rapid adoption could erode the payroll tax base that funds Social Security and Medicare while concentrating gains in firms that may pay little current income tax. But poorly designed levies could drive AI development offshore or penalize ordinary businesses for using productivity tools. The technical challenge is acute: Policymakers must define taxable activity for a technology still taking shape.

Taxing AI companies directly

Several proposals would levy taxes on the revenues, profits, or equity of firms producing frontier AI models. Sen. Mark Kelly (D-AZ) has proposed an AI Horizon Fund financed by AI company taxes. Rep. Greg Casar (D-TX) would tax large developers on token sales or revenue, with rates rising automatically as unemployment climbs.

Sen. Bernie Sanders (I-VT) has introduced legislation imposing a one-time 50% equity tax on companies with at least $200 million in AI-related receipts, payable in stock to a new sovereign wealth fund. OpenAI has separately suggested sharing 5% of its equity with the government.

Proponents see direct company taxation as the most straightforward way to redistribute AI gains. Critics warn it would penalize the firms best positioned to advance the technology and potentially violate sound tax principles by taxing revenue rather than profit. Defining which companies qualify as "AI companies" poses another challenge when most large firms now use AI in some capacity.

Taxing AI use and automation

A second category would charge fees for adopting AI or for automation-driven workforce reductions. Options include ad valorem taxes on AI purchases, levies on companies that downsize through automation, or taxes on AI tokens consumed.

Chicago already imposes a 15% tax on cloud computing services. Rep. Casar's bill would also tax companies using large models to shrink workforces, funding federal jobs programs.

Advocates argue use taxes directly address the labor displacement externality. Opponents counter that taxing automation penalizes ordinary productivity improvements that have historically raised living standards, and that such levies would disadvantage American firms against foreign competitors.

Taxing data centers and inputs

Several lawmakers have targeted AI infrastructure. Sen. Ron Wyden (D-OR) has proposed eliminating state tax breaks for data centers — currently offered in at least 38 states — and introducing a federal excise tax. Sen. Elizabeth Warren (D-MA) has suggested taxing electricity consumed by AI operations.

Virginia recently implemented a tax on electricity consumed by certain facilities, sidestepping the definition problem by metering an input rather than identifying AI workloads.

Input taxes can address environmental externalities and are difficult to avoid through offshoring. But they may be inefficient, taxing intermediate production stages rather than final outputs, and they treat clean and dirty facilities identically unless carefully designed.

Broad tax code reforms

Some economists argue the solution lies not in AI-specific levies but in removing existing distortions. The tax code currently favors capital investment over labor: employers pay payroll taxes on wages while equipment and software receive favorable treatment. This asymmetry artificially lowers the cost of automation relative to hiring.

Options include raising corporate tax rates, restoring progressive corporate taxation, reducing interest deductibility, restricting business credits like the R&D tax credit, or narrowing the gap between labor and capital treatment.

These approaches offer simplicity and durability but lack precision — they would affect many non-AI activities.

The stakes

Anthropic and OpenAI have both released policy analyses weighing tax approaches, with Anthropic warning that some mechanisms could distort investment even as they raise revenue. The concern spans the political spectrum: fiscal conservatives worry about revenue erosion while progressives focus on wealth concentration and worker protection.

The Bipartisan Policy Center evaluates tax proposals against five standards: administrative simplicity, economic efficiency, fairness, durability, and revenue-raising capacity. Few new taxes score well on all dimensions, and AI taxation faces a harder version of the problem because the activity being taxed remains in flux.

These details were first reported by the Bipartisan Policy Center in a comprehensive issue brief examining AI taxation options.