AI in Europe: Pragmatic Autonomy Amid Dependencies

Total sovereignty in AI is not considered achievable, even according to Forrester. European companies are multiplying their technological options, while Brussels invests in critical minerals and chips. Dependencies remain strong on materials, semiconductors, data centers, and models, as demonstrated by the Anthropic episode.

Forrester claims complete sovereignty remains out of reach

Forrester stated in July that total sovereignty remains beyond reach, even for China and the United States, and recommends strengthening autonomy. Dario Maisto emphasizes the importance of having options and not relying on a single supplier, especially a foreign one, to be able to migrate in case of a crisis. Sylvain Duranton from BCG believes that Europe can act on certain levers but will not be able to confine itself to the exclusive use of European technologies. According to these two experts, it is important to think in terms of degrees of sovereignty, with heightened vigilance in sectors such as military. Sylvain Duranton also asserts that creating interdependencies contributes to sovereign power. In practice, many European companies use data platforms that give them access to American, Chinese, and European models according to their needs, and for the most sensitive uses, they install the models on their own servers to limit dependencies.

China controls the majority of essential resources for the EU

According to the International Energy Agency, China accounts for about 60% of mining production and 90% of the processing of rare earth elements essential for manufacturing electronic components. The European Union represents less than 10% of global semiconductor production and does not produce any of the cutting-edge chips used for artificial intelligence. In terms of infrastructure, Europe accounts for only about 5% of global data center capacity, compared to about 75% for the United States and 15% for China, according to Andrea Renda from CEPS. In the European cloud market, AWS, Microsoft Azure, and Google Cloud together hold about 70% of the market share.

In models, the cutting edge remains outside Europe despite local players

According to a note from Rexecode, the most advanced AI technologies are currently concentrated among a few American and Chinese players, such as OpenAI, Anthropic, Meta, Google DeepMind, and DeepSeek, due to the very high development costs. However, Europe has companies like Mistral in France and Black Forest in Germany, but essential components, particularly GPUs, are mainly designed and manufactured by American companies like Nvidia and AMD, as well as by TSMC in Taiwan.

European investments and industrial advantage to reduce exposure

To mitigate these dependencies, the European Commission has set up several investment programs in critical minerals and semiconductors. Europe has an industrial advantage thanks to the Dutch company ASML, which holds the global lead in manufacturing lithography equipment for semiconductor engraving. The export of European technologies is also seen as a lever to strengthen the continent's industrial influence.

An American precedent illustrates vulnerability, and the framework is becoming clearer

In June, U.S. authorities invoked national security to compel Anthropic to restrict access to its most advanced AI models for all foreign nationals. The European Union was only able to access this technology again in September, after several months of discussions. This case arises as the United States and China hold a dominant position in the competition around artificial intelligence, while Europe seeks to reduce its dependence across the entire value chain, from electronic components to data centers to AI models. According to Dario Maisto, an analyst at Forrester, sovereignty corresponds to the absence of dependence on foreign jurisdictions and government pressures, so that no external state can prohibit access to certain technologies, data, operations, or infrastructures.