AI tax debated for UBI as IMF, Yang weigh options

What ‘taxing AI’ means and whether it can fund UBI

“Taxing AI” is a catch‑all label for several distinct bases that do not map neatly onto traditional payroll taxes. In practice, proposals cluster around: taxing AI‑related profits or excess profits; levies on compute or model infrastructure; and usage‑based charges on activity flowing through models (often framed as a token tax). Each choice targets a different part of the AI value chain and would distribute the burden across developers, cloud providers, and downstream users in different ways.

Andrew Yang’s framing argues for shifting the fiscal base away from wages and toward AI‑driven capital, with the proceeds funding universal basic income (UBI). Most expert commentary in the policy debate focuses less on eliminating labor taxes outright and more on capturing a share of AI‑related rents so that gains from automation are redistributed without stifling innovation.

Revenue realism: can an AI tax replace labor taxes?

A study by Aran Nayebi models how much AI productivity and profit capture might be needed to underwrite a broad transfer and finds that if the public captures roughly one‑third of AI capital profits, systems reaching around three times current automation productivity could sustain a UBI worth about 11% of GDP. The analysis notes that market structure is pivotal: concentrated ownership can increase taxable rents, but also heightens inequality and governance risks.

Skeptics caution that even robust AI levies may not reach the revenue scale implied by a $1,000‑per‑month benefit for every adult, as reported by The Guardian, which highlights how capital concentration complicates redistribution at national scale. That critique does not rule out material contributions from AI taxes; it underscores uncertainty about timing, incidence, and the ultimate taxable base as capabilities evolve.

Yang encapsulates the policy trade‑off succinctly: “We should stop taxing labor and tax AI instead. That can fund UBI,” said Andrew Yang. The claim is directionally clear but whether AI revenue can fully replace labor taxes depends on how much surplus AI creates, how much of it is captured by the public, and the efficiency costs of doing so.

According to the International Monetary Fund (IMF), governments weighing AI’s disruption could consider taxes on excess profits and broader capital income, and may also explore levies tied to AI usage or compute, while cautioning against designs that discourage investment. That balance, capturing rents without deterring deployment, frames the core feasibility question more realistically than an all‑or‑nothing shift from labor to machines.

Policy design choices: tax base, measurement, enforcement

Design starts with the base. Profit‑based options capture surplus where it materializes but are sensitive to accounting choices; revenue‑based options are broader but risk over‑taxing low‑margin activity; compute or usage bases are more technologically specific but must keep pace with model and hardware shifts. Dario Amodei, CEO of Anthropic, has floated what press describe as a token tax, a small percentage fee on model revenue or usage, paired with warnings that poor design could backfire by distorting incentives.

Measurement is the next hurdle. Compute levies require auditable reporting of training and inference resources; usage taxes need standardized metering of tokens or calls across providers and open‑source deployments; profit and excess‑profit regimes depend on robust transfer‑pricing and cost‑allocation rules. Policymakers also need clear definitions of what counts as AI revenue, how to treat embedded models inside larger products, and how any proceeds are routed to households or safety nets.

Enforcement is ultimately about jurisdiction and incidence. Cross‑border provision of models and cloud services heightens relocation and base‑erosion risks, while overly heavy rates could slow diffusion and reduce the very surplus that taxes aim to capture. Many fiscal discussions therefore pair AI‑specific tools with more general instruments, such as capital or excess‑profits taxation and improvements to existing safety nets, to diversify revenue sources and reduce design risk.