The EU AI Act is unfairly burdening US tech companies with compliance costs

What's this about?

People disagree about whether the EU AI Act puts unfair costs on US AI firms.

The key question is whether the law treats US firms worse than EU firms.

What supporters say

• The law can make US firms follow EU rules when people in Europe use their AI tools.
• Smaller firms may struggle more with tests, records, safety checks, and legal work than huge firms.
• US firms may use EU rules for all users, spreading these costs across their world-wide work.

What critics say

• The Act focuses on selling or using AI in Europe, not on where a firm comes from.
• One shared rule book may cost less than making a different system for every place.

How to read this

The number of points on each side does not show who is right; strong proof matters more than a long list.

The bottom line

The Act can create large costs for US firms, especially firms outside the EU.

But the evidence does not show unfair treatment based on nationality or clearly greater costs than the Act’s gains.

The EU AI Act can impose significant compliance costs on US technology companies that sell AI systems in Europe. But the available evidence does not show that US firms are unfairly singled out, or that their costs are clearly disproportionate to the Act’s benefits.

The case for

The strongest argument is that the law can reach companies far beyond Europe. It applies to firms outside the EU when their AI systems’ output is used in the bloc. General-purpose AI providers may have to prepare technical documents, explain information for downstream users, adopt copyright policies, summarize training data, test models, report serious incidents and strengthen cybersecurity. Models judged to pose systemic risks face additional duties. 1

For US companies whose engineering, legal and management operations are based elsewhere, those requirements can create substantial Europe-related costs. They may need new legal reviews, technical systems, governance procedures and reporting teams, even though they are not established in the EU.

The costs may also affect competition. Risk controls, documentation, assessments and monitoring involve fixed expenses that can weigh more heavily on smaller companies than on established technology giants. That creates a plausible risk that compliance will discourage some new entrants or slow innovation. However, existing research does not show that US companies are treated worse than EU companies of similar size and risk.

The rules could also spread beyond Europe. Multinational companies may choose to use one global product standard rather than build separate systems for European and other markets. Research on the EU’s wider regulatory influence supports this possibility, meaning US firms could face costs across their global operations, not only in products aimed at European users. 3 But that would demonstrate the EU’s regulatory reach, not discrimination based on nationality.

Some costs may also be lower than the law appears to impose at first glance. Large US AI companies already conduct safety tests, evaluations, documentation and security work that overlap with the Act’s requirements. They would still face added expenses from formal paperwork, audits, standardized disclosures and other controls, but the true extra cost depends on what systems and compliance teams they already have.

The case against

The Act’s basic design is based on market activity and risk, not nationality. Its requirements depend on a provider’s role, the use of its AI system, its risk category and its connection to the European market. EU and non-EU companies operating in the same regulated category can face the same duties. The fact that major US model developers have extensive obligations therefore does not, by itself, prove unequal treatment. 4

The law may also replace a patchwork of national rules with one framework for the EU’s single market. A common set of requirements could reduce the need for companies to meet different standards in different European countries. 5 Trust and safety rules may bring short-term organizational costs, but they could also reduce risks, increase user confidence, support adoption and make investment decisions more predictable.

Those possible benefits have not been measured specifically for the EU AI Act, however. Nor is there reliable company-level evidence showing how much the law adds to the costs of particular US firms compared with comparable European firms. Preliminary reviews suggest that some existing industry practices overlap with the Act, but they do not provide independently audited accounts.

The bottom line

The evidence moderately supports a narrower version of the claim: the EU AI Act can impose material, and sometimes globally spread, compliance costs on US technology companies. The legal exposure is well established.

But the stronger claim—that the costs are unfairly disproportionate to those faced by comparable EU firms or to the Act’s public benefits—is not established. The evidence supporting that conclusion is much weaker than the evidence showing that US firms fall within the law’s reach. Key uncertainties remain over company-level costs, the value of avoided harms and other benefits, and how regulators classify and enforce the rules.