AI is changing the economics of the internet, and with it the old assumption that technology can be left to “work itself out.” For decades, policymakers largely treated new digital tools as a kind of economic weather: disruptive, sometimes chaotic, but not something that required immediate, targeted fiscal rules. That hands-off model made sense when the main question was whether platforms could scale faster than regulators. Today, the question is different. AI systems are not just adding convenience or automating tasks at the margins; they are reshaping how value is created, captured, and distributed—often across borders, often through opaque supply chains, and increasingly through infrastructure that is partly public and partly privately owned.
In that context, a growing policy argument has emerged: if AI is going to “pay its way,” the tax system should be updated so that the benefits AI extracts from markets, data, labor, energy, and public goods are reflected more directly in government revenue. The proposal is not simply about raising money. It is about aligning incentives, improving fairness, and building a framework that is enforceable enough to survive the next wave of innovation.
The core idea sounds straightforward—tax it properly—but the details are where the debate becomes genuinely interesting. What does “properly” mean when AI is both a product and a general-purpose capability? When the “user” might be a company deploying a model, while the “producer” might be a cloud provider, a model developer, a data aggregator, or a compute supplier? When value can be created in one country, trained in another, served from a third, and monetized globally? And when the biggest economic effects may show up not as new profits, but as changes in bargaining power, wage dynamics, and market concentration?
A post-laissez-faire internet is not a slogan. It is an admission that the old playbook—wait, observe, regulate later—doesn’t match the speed and scale of AI-driven change. The tax system, in this view, is one of the few levers governments have that can be both broad and precise enough to influence behavior without micromanaging technology.
Why the “hands-off” model is losing credibility
The laissez-faire approach worked tolerably well for earlier internet technologies because the economic footprint of those technologies was easier to map. Advertising-based platforms, for example, had clear revenue streams and identifiable business models. Even when enforcement was difficult, the basic unit of taxation—profits, sales, payroll—was familiar.
AI complicates that mapping. Consider what happens when an AI system is deployed inside a business workflow. The company may pay for access to a model, but the economic gains can be distributed across multiple layers: reduced headcount in some roles, higher throughput in others, improved pricing power, faster customer acquisition, and lower error rates. Some of these gains become visible as profit. Others appear as cost savings that don’t automatically translate into taxable income in the short term. Meanwhile, the costs of AI—compute, data acquisition, engineering talent—are often front-loaded, which can create temporary losses or low margins that do not reflect the long-run value being extracted.
This is why critics argue that “tax it properly” is not just about catching companies that avoid taxes. It is about updating the tax base so it corresponds more closely to the economic activity AI generates. If the tax system continues to rely on outdated assumptions about where value is created and how it is measured, governments risk under-collecting revenue while also failing to address distributional harms.
The argument is also political, though it is framed as technical. When AI reduces demand for certain kinds of labor, increases productivity unevenly, and concentrates market power among firms that can afford compute and data advantages, the social contract strains. Tax policy becomes part of the response: either governments capture a fair share of the gains to fund adjustment programs, or they allow the gains to accumulate privately while the costs are socialized.
What “AI pays its way” actually means
“Pay its way” can sound like a moral claim, but in policy terms it usually translates into three practical goals.
First, revenue adequacy: governments need stable funding for public services as AI changes the economy. If AI shifts profits into structures that are hard to tax—or if it generates value without producing taxable income—public budgets face pressure.
Second, fairness: if AI firms benefit from public infrastructure (research ecosystems, legal systems, education pipelines, energy grids, telecom networks) and from public goods (data collected under regulatory frameworks, public procurement, public safety), then it is reasonable to ask whether the tax system reflects that shared foundation.
Third, behavioral alignment: taxes can shape incentives. If AI deployment is socially beneficial, the tax system should not discourage it. But if AI creates externalities—such as displacement without adequate worker transition, increased energy demand without corresponding contributions, or market power that harms competition—then taxes can help internalize those costs.
The “simpler option” framing suggests that rather than inventing entirely new institutions, policymakers can adapt existing tax concepts to AI’s realities. That might include revisiting corporate tax rules, refining withholding and nexus standards for cross-border digital services, and creating targeted mechanisms for AI-specific value capture—without turning every model deployment into a bespoke regulatory case.
Where the value is: data, compute, and the invisible supply chain
One reason AI taxation is hard is that AI is not a single product with a single supply chain. It is a stack.
At the bottom is compute: GPUs, electricity, cooling, data center capacity. Compute is tangible, measurable, and increasingly concentrated in regions with favorable energy and infrastructure. But compute alone doesn’t explain why AI can generate outsized economic value. The real leverage comes from training and inference at scale, from the ability to integrate models into workflows, and from the data and feedback loops that improve performance.
Then there is data. Data can be proprietary, scraped, licensed, or derived from user interactions. Some data is personal, some is non-personal, and some is generated by systems themselves. The legal and ethical status of data varies widely, which makes it difficult to build a tax mechanism that depends on data provenance without creating loopholes or perverse incentives.
Finally, there is distribution and integration. A model might be trained by one firm, hosted by another, and used by a third. The firm that captures revenue might not be the firm that created the most value. In many cases, the “value capture” is tied to who controls the interface to customers and who owns the relationship with the end user.
This is where “tax it properly” becomes less about punishing any one actor and more about designing a system that can follow value across the stack. If the tax system only taxes profits where they are booked, it will miss value created elsewhere. If it only taxes revenue, it may penalize firms that invest heavily upfront. If it only taxes payroll, it may fail to capture gains from automation and substitution.
A credible approach needs to balance these tradeoffs.
The policy proposals behind the phrase
While different countries and think tanks use different language, the underlying proposals tend to cluster around a few themes.
One theme is updating nexus and source rules for AI-related services. Traditional tax systems assume that a company has a physical presence where it earns income. Digital businesses often challenge that assumption. AI intensifies the challenge because models can be served remotely, and value can be generated through interactions that leave little physical footprint. Policymakers argue that AI should not be able to monetize a market without contributing to that market’s tax base.
Another theme is rethinking how corporate tax interacts with intangible assets. AI models are intangible, but so are software, patents, and brand value. The difference is that AI’s economic impact can be faster and more diffuse. Firms may shift profits through transfer pricing arrangements involving IP, licensing, and service fees. Tax authorities worry that current rules allow too much flexibility for multinational groups to allocate profits in ways that do not reflect economic substance.
A third theme is targeted levies tied to specific externalities. If AI increases energy consumption dramatically, some propose energy-linked contributions. If AI accelerates labor displacement, some propose funding mechanisms for worker transition. If AI increases market concentration, some propose competition-linked fiscal measures. These ideas are controversial because they require measurement and can be gamed, but they also reflect a broader truth: AI’s costs are not purely financial.
A fourth theme is “minimum taxation” or floor mechanisms to prevent base erosion. If AI firms can reduce taxable income through deductions, intercompany payments, or jurisdictional arbitrage, then minimum taxes can help ensure that some portion of economic gains is captured regardless of where profits are booked. This is less about AI being special and more about AI being a stress test for the tax system’s ability to handle modern multinational structures.
The unique twist: taxing AI as a governance tool, not just a revenue tool
What makes the current debate feel different from earlier digital tax discussions is the emphasis on governance. The question is no longer only “How do we collect taxes from tech companies?” It is “How do we design a fiscal framework that shapes the trajectory of AI adoption?”
In practice, that means policymakers are looking for taxes that can do more than raise revenue. They want taxes that can influence investment patterns, encourage responsible deployment, and reduce the likelihood that AI benefits accrue without corresponding contributions to social stability.
For example, if a tax system rewards firms for maximizing short-term profits while ignoring long-term societal costs, AI deployment may accelerate even when it produces net harm. Conversely, if taxes are designed to reward transparency, workforce development, and compliance with safety and accountability standards, then fiscal policy can become part of the incentive structure that determines whether AI is deployed responsibly.
This is where “tax it properly” becomes a kind of shorthand for a broader package: tax rules that are consistent, enforceable, and aligned with the real economic and social effects of AI.
The measurement problem—and why it shouldn’t be ignored
Skeptics argue that AI taxation risks becoming a bureaucratic exercise. How do you measure “AI value” when AI is embedded everywhere