European AI – The Language Effect

The EU is down but not out.

• With 452m relatively well-heeled users who do not speak English as their first language, Europe is ripe for homegrown AI offerings, meaning that there is still opportunity as soon as the EU can get out of the way of the AI industry.
• Europe, and the EU in particular, has 24 official languages as well as a wide diversity of cultures, which means that LLMs trained on English-based content and culture are unlikely to be ideal for these users even when perfectly translated.
• This, combined with a general feeling that Europe is already too dependent on US technology, has created an environment that is ripe for homegrown technology to be welcomed with open arms.
• Unfortunately, very little has been created so far, with just one LLM provider (Mistral) widely available and a few other offerings such as ElevenLabs, Black Forest Labs and ChapsVision, which is an AI software offering that connects and analyses data in a similar way to Palantir.
• The scale of these offerings also remains very small, as the leading player, Mistral, has a target of 200MW of capacity and an annual revenue run rate of $1.1bn, compared to Anthropic, which is already more than 60x its size.
• Nvidia has already made a big push to jolt Europe out of its stupor, and despite few results to date, it has not given up and will host a European GTC event in Berlin this week.
• This is because there are some signs that the regulatory situation is improving, although it is clear to me that the EU has not gone nearly far enough.
• Enforcement of the regulations that have constrained the LLM providers has been pushed back to the end of 2027, but a lot more is needed.
• I think that the AI developers and operators of AI services require the AI Act to be changed, and delaying rules is not going to give them the confidence to invest in Europe.
• However, there has been explicit acknowledgement from EU policymakers that the AI Act goes too far and is hindering the development of AI in Europe, which is a sign that the environment is likely to improve.
• I think that it is very unlikely that we will see big investments before the regulatory environment changes, and in these sorts of things the EU can be glacial in its speed of action.
• Hence, Europe continues to be a minnow in the world of AI, with the Middle East taking a lot of the lost momentum despite the conflict that is still smouldering in its backyard.
• LLMs trained on Arabic-specific content give much better responses for Arabic natives, which is a strong sign that the same will be true for all of the 24 European languages.
• Hence, the opportunity for Sovereign AI to offer localised intelligence for citizen services, healthcare and the public sector continues to exist all over Europe, but when it will materialise is less clear.
• The net result is that the ball is in the EU’s court, and I think the minute that things change, Europe will spring to life and grow much faster to catch up with the USA and China.
• This will create space for many more companies as Mistral clearly cannot cover the entire continent on its own, and there will also be requirements for different silicon offerings as well as a jump in demand for electricity.
• Jensen clearly has much more faith in the EU regulators than I do, but one can but hope that the EU decides that losing the greatest innovation of 2026 (OpenClaw) to the Americans was one insult too many.