Who should pay for AI?

AI is embedded in everyday technologies at low or no cost to most users, but it isn’t free. As AI proliferates, so do its demands on electricity and water resources, along with broader societal challenges such as labor market disruption, privacy concerns, environmental degradation, and the spread of misinformation. Some of these hidden costs are already landing on the public. These costs are on the rise, but affected communities may not be on the hook forever.

New research from Carnegie Mellon University’s Department of Engineering and Public Policy (EPP) offers a dependable way to mitigate these costs through the tax code. In recently published work, Juliette Faivre, an incoming Ph.D. student, and Sarah H. Cen, assistant professor of EPP and electrical and computer engineering analyze how targeted taxation could promote more socially responsible AI operations.

They point out a fundamental inequity: the appetite for AI is broad and diffuse, but its consequences burden specific, localized communities. Neighborhoods that host data centers have experienced water supply interruptions. AEP Ohio, the power company serving Ohio’s data center corridor, told customers to expect monthly electricity bills to climb by $27—a rate hike attributed in part to the soaring needs of the nearby tech facilities. Residents experience utility strain and increasing cost regardless of personal use.

The extraction of human-produced data for AI training presents another significant inequity. Most creative contributors never consented to, nor received compensation for, the use of their work for profit-driven AI firms. Furthermore, creative professionals face reduced demand for their labor as AI systems get better at mimicking their creative voices.

The researchers propose that taxation can balance the harmful effects of AI in three ways: 1) corrective: disincentivizing harmful activity such as overconsumption of water or electricity through price signals; 2) redistributive: directing tax revenue toward restorative or remedial action; and 3) regulatory capacity: using tax revenue to fund AI oversight through monitoring and enforcing regulation.

AI taxation deserves closer consideration by policymakers, as it could add the missing public-finance piece of AI governance: how to pay for the needs an AI economy creates.

Juliette Faivre, Incoming Ph.D student, Engineering and Public Policy

While other approaches to mitigating the costs of AI have been explored, the researchers note significant barriers to their enforceability. Proposals for risk-management frameworks, transparency obligations, incident reporting, and others require a complicated evaluation infrastructure that is mostly untested. Mitigation taxes (often referred to as Pigouvian taxes), on the other hand, have been successfully used for decades in the U.S. to offset societal costs of private industry goods and behaviors.

Taxation is an appealing strategy because it leverages a proven mechanism for managing market externalities. Moreover, because even small businesses possess a basic tax compliance framework, they can swiftly adapt to new obligations, making timely deployment feasible. Crucially, a public tax framework also allows for independent oversight.

The research highlights another built-in advantage: in the U.S., tax laws can be passed through budget reconciliation. This process bypasses the filibuster, allowing Congress to avoid typical gridlock.

AI taxation is no longer theoretical. In the Senate, proposals to address AI-driven harms have been introduced, and some industry leaders have recently supported AI taxation discussions.

Faivre and Cen are careful to point out that AI taxation is not a magic bullet. Firms may attempt to avoid taxes by moving to jurisdictions with lower taxation or weak regulatory enforcement. Taxation may not be an appropriate remedy for some harms, such as privacy violations and discrimination. Finally, taxation always carries the risk of chilling innovation, threatening global competitiveness. The researchers emphasize that taxation should be used to complement other AI governance tools rather than as a standalone fix.

“AI taxation deserves closer consideration by policymakers, as it could add the missing public-finance piece of AI governance: how to pay for the needs an AI economy creates,” says Faivre.

By shifting a portion of the financial burden back onto the industry, a well-designed tax framework can help align tech’s rapid growth with the public benefit. While not a comprehensive solution, integrating taxation into the broader regulatory toolkit offers a pragmatic way to restore equity to the communities currently subsidizing AI’s footprint.