The EU AI Act’s Costs to American Innovation

The EU AI Act governs the deployment and use of artificial intelligence (AI) in the European Union (EU), but its scope extends beyond European borders. It binds any provider worldwide whose AI models or systems reach users in the bloc. For U.S.-based tech firms—the leading suppliers of both models and systems to the EU—that means significant compliance costs and potential fines. This could deter American investment in AI and make it harder for the United States to maintain its technological edge over China.

The Act regulates AI applications by risk tier. It bans systems it deems to have unacceptable risk, such as social scoring, and imposes light rules on limited-risk systems and minimal rules on low-risk systems. The strictest compliance requirements fall on high-risk AI systems (HRAIS).

The Act also regulates general-purpose AI (GPAI) models, the models on which developers build AI systems. GPAI providers face extensive regulatory obligations, including preparing transparency documentation and training data summaries, abiding by EU copyright law, and providing detailed information to downstream developers integrating the model into their own systems, so they can do so responsibly.

Because U.S. firms are major suppliers of GPAI models and HRAIS systems in Europe—which together carry most of the Act’s compliance burden—American firms will bear a disproportionate share of the Act's compliance costs.

Systems that carry out critical functions in sectors such as health care, critical infrastructure, education and training, employment, and law enforcement are classified as high-risk under the Act, and across the EU, these systems are largely American-built. Palantir, a U.S. company, supplies law enforcement AI software in numerous European countries, including Denmark, Germany, and the Netherlands. In health care, many organizations rely on U.S. technology firms for both digital infrastructure and foundation models themselves. Of the 20 key players identified in one industry report, 65 percent are U.S.-headquartered, including Microsoft, Google, Oracle, NVIDIA, and Amazon Web Services.

American dominance extends to GPAI as well. The 2025 Stanford AI Index report recorded 40 U.S.-developed foundation models compared with 3 European-developed models, and the Draghi report found that 70 percent of all foundation models produced since 2017 originated in the United States. Europe’s dependence on U.S.-developed AI is also visible in downstream applications. U.S.-developed chatbots, for instance, capture more than 95 percent of European traffic.

Within GPAI, a subcategory of especially powerful models—those trained above 10²⁵ FLOPs—faces stricter obligations as models with "systemic risk." Of the 12 developers whose models have likely crossed that threshold, 7 are American and only 1 is European.

The Act's penalty structure can make the financial exposure of a violation far larger than the revenue of the AI product involved. Fines run up to 7 percent of a company's global annual turnover, with lower thresholds for less serious violations. In every case, the ceiling is calculated against worldwide revenue, not EU revenue—disadvantaging U.S. firms given their global footprint.

For several U.S. tech firms, AI accounts for only a small share of their revenue, yet the Act calculates fines based on a firm’s total turnover, rather than AI-specific revenue. For diversified American companies, that means revenue from entirely unrelated businesses can drive up the size of the fine.

Microsoft's annual revenue for fiscal year 2025 was $281.7 billion, while its projected annual revenue from its AI business stood at $13 billion. This figure includes revenue from AI infrastructure and AI models, with only the latter falling under the scope of the EU AI Act. The regulated slice––where any non-compliance could arise––accounts for less than 5 percent of the company's turnover.

The United States and China are locked in intense technological competition, and as their AI model performance gap narrows, sustained investment in innovation is essential to U.S. leadership. Large turnover-based fines would leave U.S. firms with fewer resources for that investment, while discouraging them from pioneering innovations for fear of heavy fines.

Chinese AI firms, by contrast, face no comparable burden, given their smaller footprint in the European market. In 2024, it produced 14 foundation models—less than half the U.S. total—and Chinese chatbots capture just 0.03 percent of Europe's chatbot market. In Germany, only 2 percent of AI-using companies rely on DeepSeek and 1 percent on Qwen, compared with more than 76 percent that rely on American models. Without comparable compliance costs, Chinese firms can instead direct more resources toward innovation, expand into contested markets in other parts of the world—including Africa, Latin America, and Southeast Asia—and catch up with the United States.

The EU AI Act's complex compliance burden risks eroding the competitiveness of U.S. firms that are key drivers of global AI innovation. Europe should instead simplify its regulatory framework and deepen collaboration with Washington on AI policy––a move that serves Europe's own interests by keeping American investment flowing and its domestic ecosystem flourishing. The shared goal should be to bolster innovation and drive economic growth on both sides of the Atlantic.