Taxation of artificial intelligence: why governments cannot define how to collect

Governments around the world are determined to tax artificial intelligence, but they face a theoretical obstacle: no one can define exactly what should be taxed. While US technology companies and legislators present generic proposals, public policy makers struggle to translate AI complexity into something measurable for the tax.

Why is AI taxation so difficult to conceptualize?

The central difficulty lies in the intangible and multifaceted nature of AI. Unlike a physical product or a clearly delimited service, artificial intelligence can be viewed as software, as a service (SaaS), as intellectual property, as manpower automation or as computer infrastructure. Each of these perspectives would suggest a different type of taxation — value added tax, royalty income tax, contribution to automation or data use tax. The absence of a single legal definition makes it difficult to create a uniform rate or basis of calculation.

What has been proposed so far in the United States about taxation?

In the United States, the debate is at an early stage. Some legislators suggested a "bot tax" that would focus on companies that replace human workers with automated systems. Other proposals aim at the revenue generated by AI models, treating it as a digital service. However, according to Bloomberg Law, these ideas are vague and lack technical detail — they do not specify, for example, whether the fee would fall on the model developer, the end user or the infrastructure provider.

How do other countries deal with the challenge of AI taxation?

The European Union already discusses a tax on digital services that could be extended to the AI, but faces resistance from Member States. In Brazil, the theme is still incipient, but the growing adoption of AI systems in sectors such as finance and health should pressure the government to position itself. A reflection of this is the debate on the energy cost of data centers, which has already generated controversies in the US on dealership profits. Technical complexity is added to fiscal urgency: many governments see in AI a new source of revenue to compensate for the automation of traditional jobs.

What are the main practical obstacles to AI taxation?

Besides the lack of definition, there are measurement challenges. How to calculate the value generated by a trained language model with billions of parameters? How to separate commercial use from personal use? How to tax companies that use AI indirectly, as in recommendation systems or chatbots? Experts point out that any tax attempt will require clear rules on transfer pricing, royalties and digital jurisdiction, issues that are already thorny in traditional electronic commerce.

What is at stake for the AI ecosystem with taxation?

The course of taxation can directly impact innovation. Companies such as Anthropic, which recently launched connectors and workflows for small businesses, depend on a predictable regulatory environment. Similarly, Scale AI and other startups that provide data and assessments for models need to know whether their services will be taxed as software or as data services. A poorly designed tax burden can discourage investment or simply be evaded via corporate restructuring.

What is the most likely scenario for AI taxation?

Analysts heard by Bloomberg Law believe that the solution will not come from a single tax, but from a combination of mechanisms: taxation on profits from companies that use AI intensively, fees on the use of personal data for training, and social contributions to compensate for lost jobs. The consensus is that the process will be slow and fragmented, with countries testing different approaches before possible global coordination, similar to what happened with OECD Digital Services Tax.

Frequently Asked Questions about AI charging

What exactly do governments want to tax on AI?

The idea is to tax the economic value generated by AI systems, whether in the form of revenues from companies that develop or use them, or in the extra profit obtained with automation. The difficulty is that this value is difficult to measure and assign to a specific stage.

How can AI taxation affect Brazilian startups?

Startups using AI as a competitive differential can face increased costs if taxation falls on data or models. On the other hand, clear rules can bring legal certainty to investors.

Is there a country that has already implemented the AI tax?

No country has yet a specific AI tax in force. Advanced discussions take place in the European Union and in some American states, but none have been approved so far.