The future of tax policy: A public finance framework for the age of AI

As artificial intelligence transforms our economy, policymakers worldwide are grappling with how to adapt our systems of taxation and public finance for an automated future. Common proposals—which we explore in more detail below—range from taxing robots and computing power to levying fees on AI-generated tokens and digital services. Yet without a coherent framework for evaluating these options, we risk implementing policies that could hinder innovation and undermine competitiveness while failing to address the fundamental fiscal challenges ahead.

Our recent research provides a framework for addressing these challenges by examining how taxation systems should evolve as AI transforms production and employment. We find that timing is key: Certain reforms make sense now, as AI is starting to displace labor, that would complement innovation and economic growth, while others could undermine efficiency and would be counterproductive until AI systems become far more autonomous. Understanding this distinction is crucial for policymakers seeking to manage the economic transition and maintain fiscal sustainability while fostering the innovation that will drive future prosperity.

The coming fiscal challenge

The modern tax system in the U.S. rests on two pillars: labor income and, to a lesser extent, consumption. According to 2023 data from a Congressional report, about three quarters of all U.S. federal tax revenue comes from labor.¹ Unlike in many other advanced countries that have extensive value-added tax (VAT) systems, consumption taxation only plays a minor role at the federal level in the U.S., but it plays a significant role in the form of sales taxes at the state level. AI threatens to erode the first pillar—taxes on labor—by reducing demand for human labor across many occupations. While the extent and timing remain uncertain, even modest labor displacement could significantly strain public finances at a time when funding for social safety nets may be needed most.

This challenge is not merely theoretical. Labor’s share of income has already declined in recent decades, and many economists expect AI to accelerate this trend. It is not clear yet, but empirical evidence is emerging that recent disappointing job data may be AI-related. This may be the beginning of a more significant trend of labor displacement. As machines perform an expanding range of tasks, from customer service to complex analysis, the traditional tax base of wages and salaries may shrink dramatically. Policymakers must prepare for this possibility while recognizing that the speed and scope of the transition remain uncertain.

A framework for taxation in the age of AI

To evaluate AI tax proposals effectively, we must consider the technology’s likely economic impact in stages. The first stage—potentially already beginning—involves AI systems that enhance productivity while gradually displacing workers. During this phase, humans remain the primary consumers and beneficiaries of economic output, but their role in production diminishes. Capital owners capture increasing income shares, potentially exacerbating inequality.

A second stage might see AI systems both producing and absorbing an increasing share of the economy’s resources—building data centers, power plants, and supporting infrastructure—in a manner that generates little labor income or human consumption. For example, if AI systems become sufficiently intelligent, they might be able to run AI companies by themselves while reinvesting most surplus and therefore generating little taxable profit and, all in all, little tax revenue under our current tax system. While this scenario is only a hypothetical, considering it helps stress-test our fiscal frameworks.

Approaching our central question in terms of stages of AI advancement underlines why timing matters for AI taxation. Policies appropriate for an economy where humans direct all capital investment may prove inadequate once AI systems operate more autonomously. By understanding these dynamics, we can implement policies that work today while maintaining flexibility for tomorrow.

Making sense of AI tax proposals

Against this backdrop, how should we evaluate the proliferating proposals for AI-related taxes? The crucial distinction is between taxing AI services at the point of consumption versus taxing the underlying capital assets. This difference determines whether a tax supports or undermines innovation.

Consider five common proposals:

Digital services taxes on AI-provided services to consumers represent consumption taxation. When governments tax streaming services, cloud storage, or AI assistants at the point of purchase, they follow sound economic principles. These taxes can integrate into existing consumption taxation systems without deterring business investment in AI and infrastructure. Critically, they allow companies to build cutting-edge AI systems while capturing value where it’s created.

Token taxes on AI-generated content function as consumption taxes when applied to final users. A fee per thousand words or images generated makes economic sense for consumer applications. However, such taxes must be oriented at the retail level—like current sales taxes—and exempt business-to-business transactions to avoid cascading effects that would make AI products less competitive. The goal is to capture revenue without creating barriers to AI adoption in production processes.

Robot services taxes on automated services like robotic delivery or AI customer support also represent consumption taxation. Rather than taxing the robots themselves, these levies apply to services robots provide to end users. This approach generates revenue while preserving incentives for businesses to invest in automation that enhances productivity—a key requirement for maintaining technological leadership.

By contrast, robot taxes on owning or operating robotic equipment would directly tax productive capital. Such taxes would discourage precisely the investments needed to build AI infrastructure. Such handicaps on automation investment could hinder productivity growth and the broad economic gains it enables.

Similarly, compute taxes on computational resources or AI hardware would discourage data center development. Taxing the fundamental infrastructure of AI development would be like taxing steel during the industrial revolution—a self-defeating policy that could slow the productivity growth needed to fund public priorities.

The innovation-friendly path forward

As labor income taxation becomes less viable, the primary lesson is that consumption taxation must carry more fiscal weight. This shift aligns naturally with pro-innovation policies, as consumption taxes avoid penalizing the capital investments essential for future innovation. In addition to this fundamental shift to focus more on consumption, we identify several other considerations that should guide this transition.

First, governments must not only expand but also modernize consumption tax systems to handle digital services effectively. Achieving technological leadership requires tax frameworks that can efficiently process international digital transactions without creating compliance burdens that drive business elsewhere.

Second, the declining importance of labor income changes optimal tax design. Currently, governments keep consumption taxes relatively uniform to avoid compounding labor market distortions. But as labor becomes less central, there’s more scope for differential consumption taxation based on practical considerations like ease of evasion or administration costs. For example, policymakers might tax easily monitored digital services more heavily than goods prone to black market activity.

Third, inequality concerns won’t disappear—they may in fact intensify. While consumption taxes are often criticized as regressive, it is possible to tax consumption progressively (e.g., by taxing the difference between income and net savings at progressive rates), and consumption taxes become more important for funding programs that ensure broad-based prosperity when capital ownership is highly concentrated. The solution isn’t to avoid consumption taxes but to design the overall tax and transfer system to offset any negative distributive effects of AI.

Fourth, identifying and taxing economic rents becomes more valuable as labor’s share declines. Rents—returns above what is necessary to induce an activity—can be taxed without distorting investment decisions. In an AI-driven economy, sources of rent may include unimproved land, spectrum rights, unique datasets that cannot be replicated, and monopoly profits from market concentration in AI. Corporate income taxation can serve as one vehicle for capturing such rents, provided it is designed with appropriate deductions for depreciation and interest payments that exempt the normal return to capital while still reaching excess returns. While distinguishing true rents from normal returns remains challenging in practice, building capacity to identify and tax such rents provides a non-distortionary revenue source that complements consumption taxation.

Managing the fiscal implications of labor displacement requires careful planning. As AI reduces demand for certain jobs, government revenues from payroll taxes as a fraction of GDP will decline just as needs for retraining programs and transition support increase. Consumption taxes can help bridge this gap by maintaining revenue streams even as employment patterns shift.

Implementation priorities

Translating these principles into practice requires attention to several factors:

Distinguishing intermediate from final use prevents taxes from cascading through production chains. Token taxes should apply to consumer chatbots but not AI systems used in manufacturing or research. This distinction preserves incentives for investment and innovation while still capturing revenue.

Administrative feasibility must guide implementation. While economic theory might suggest optimal tax rates for different AI services, practical enforcement capabilities constrain what’s possible. Simple, transparent systems are more friendly for business and innovation than complex frameworks that create uncertainty.

International coordination becomes crucial, as AI services cross borders effortlessly. The nations leading in AI development could drive efforts to harmonize digital taxation frameworks that prevent both tax avoidance and double taxation. This supports their technological leadership by ensuring robust tax frameworks become models for international adoption.

Preparing for an uncertain future

While focusing on near-term adaptation, prudent policy must consider more transformative scenarios, for example, the hypothetical (but not implausible) future in which an artificial general intelligence (AGI) is able to operate as an independent firm. In that case, traditional taxation of human labor income or consumption might prove insufficient.  Governments might need to tax the capital accumulation of AGI systems directly, similar to how they currently tax corporations, although the tax would need to be broader to cover all of the AGI’s resource accumulation rather than just accounting profits. Our companion technical paper develops a formal framework for this scenario, showing that the optimal tax rate on autonomous AGI entities depends crucially on human time preferences (humans’ willingness to postpone consumption)—a striking result that underscores how human values remain central to policy design even in an AI-dominated economy.

Given the radical uncertainty surrounding AI’s trajectory, prudent policy suggests maintaining flexibility in tax frameworks and avoiding commitments that would preclude future adaptation to changed circumstances. Equity-based mechanisms deserve particular attention as complements to tax reform. Sovereign wealth funds that invest in AI companies can capture returns that automatically scale with AI’s economic impact. Windfall clauses—voluntary commitments by AI companies to share gains broadly if they achieve transformative breakthroughs—provide insurance without requiring new taxes. Policies ensuring broader ownership of AI enterprises would address inequality at its source rather than relying solely on after-the-fact redistribution. These approaches offer fiscal insurance: If AI development stalls, returns remain modest; if AI transforms the economy, public participation in the upside is secured.

Conclusion

Maintaining leadership in the AI era requires ambitious fiscal innovation. By shifting from a primary reliance on labor taxation to consumption taxes, policymakers can generate revenue without hampering innovation. The framework presented here offers a path to finance government during the AI transformation while maintaining competitiveness.

Our key recommendations are:

1. Shift from taxes on labor as the primary source of revenue to taxes on consumption.
2. Modernize consumption tax systems for digital and AI retail services.
3. Avoid taxing AI capital assets in the short term to avoid distorting infrastructure development.
4. Build administrative capacity.
5. Maintain flexibility for future adaptation of a tax on AI-related resource accumulation if AI entities become the primary drivers of value creation in the economy.

The fiscal challenges of an AI-driven economy may soon become tangible, but proper planning can help prepare for them. By adapting proven principles of public finance to new circumstances, we can maintain fiscal sustainability while ensuring that the gains from AI are broadly shared. The choice is clear: Design tax systems that harness AI’s potential for broad-based prosperity, or watch fiscal frameworks buckle under technological change.