Level : Intermediate
European AI: where does Europe stand against the US and China?
By The Info Agent IA editorial team ·
European AI is real, led by Mistral AI, but Europe trails the United States by a wide margin on money invested and on top-performing models, and trails China on the number of notable models. Its strengths lie elsewhere: fast-growing adoption by businesses, a protective legal framework and services that keep your data on the continent.
Who are Europe’s AI players?
“European AI” covers two quite different jobs. First, model makers, who design and train the engines behind AI tools. Second, companies that build services on top of those engines: assistants, browsers, automation tools. Europe has few names in the first group and many more in the second.
Model makers. Mistral AI, founded in Paris in April 2023 by Arthur Mensch, Guillaume Lample and Timothée Lacroix, is the only European maker of large general-purpose models among the site’s profiles. It releases some of its models under the Apache 2.0 licence and runs the Le Chat assistant. In Warsaw, AionLabs adapts other companies’ open models for storytelling and role-play rather than training its own from scratch.
One special case is Google DeepMind. DeepMind was founded in London in 2010 and is still headquartered there, but it belongs to Alphabet, an American group. Calling it “European” would be misleading.
Services and tools. Several European companies offer assistants or AI agent tools without building their own large models.
| Company | Country | What it offers | Models used |
|---|---|---|---|
| Mistral AI | France | Models, API, Le Chat assistant | Its own |
| AionLabs (Deep Forge) | Poland | Storytelling models via API | Open bases (DeepSeek, GLM, Llama) |
| Proton (Lumo) | Switzerland | Privacy-focused assistant | Open models (Qwen, GLM, Apertus) |
| Opera (Neon) | Norway | Browser with agents | OpenAI, Anthropic and Google models |
| Make (Celonis group) | Czech Republic | No-code automation with agents | Third-party providers |
| n8n | Germany | Automation with AI building blocks | Third-party providers |
Sources: the site’s profiles, checked on 22 October 2026. “Europe” is used broadly here: Switzerland and Norway are not EU members.
The pattern is clear. Apart from Mistral AI, European value sits in applications, privacy and integration rather than in building frontier models. For the global picture, see our guide who builds AI?.
Money: a huge gap
The first gap shows up in private investment, meaning money from funds and companies.
| Region | 2024 | 2025 |
|---|---|---|
| United States | $94.6bn | $240.6bn |
| Europe | $16.9bn | $17.6bn |
| China | $8.1bn | $10.4bn |
| World | $130.9bn | $290.1bn |
Private investment in AI, in billions of US dollars. Source: Our World in Data, “private investment in artificial intelligence” series, checked on 22 October 2026.
Read these numbers with care. They only count private money, not public spending, so China’s low total on this measure says nothing about its overall effort. Nor do they tell you what it actually costs to train a model. More series are gathered on our AI in numbers page.
Models: China far ahead of Europe
The second indicator is the number of AI models that the research institute Epoch AI rates as “notable”: models that advanced the state of the art, were widely cited or widely used. According to Epoch AI (checked on 22 October 2026), 62 notable models released in 2025 came from the United States and 34 from China, against a single one from the United Kingdom. All other countries apart from South Korea and Canada, France and Germany included, add up to just 2. For 2026, at the time of checking: 49 in the US, 29 in China, none in the UK and 1 for all other countries combined.
On performance, no European model sits near the top of the ranking based on Epoch AI’s tests. Mistral AI’s best-rated model, Mistral Large 4, released on 6 October 2026, scores 79.7 out of 100. The two leaders, Anthropic’s Claude Sonnet 5.5 and Claude Opus 5.5, reach 95.9.
- Claude Sonnet 5.595.9
- Gemini 4 Argon95.3
- DeepSeek V4.1 Flash 101790.6
- Kimi K2.684.4
- Mistral Large 479.7
That deserves some nuance. The index measures maths, science, coding and agent tasks, not how well a model writes in French or German. Price matters too: according to models.dev, as of 22 October 2026, Mistral Large 4 costs $0.68 per million input tokens and $2.09 per million output tokens, against $2 and $10 for Claude Sonnet 5.5. To see what these scores are really worth, read our article understanding AI benchmarks and their pitfalls.
Usage: Europe is speeding up
Adoption tells a more encouraging story. According to Eurostat (checked on 22 October 2026), the share of EU businesses with 10 or more employees using at least one AI technology rose from 7.65% in 2021 to 19.95% in 2025. Gaps between countries remain wide: 42.03% in Denmark, 25.97% in Germany, 20.27% in Spain, 18.16% in France and 16.4% in Italy in 2025.
Among individuals, the share of 16 to 74-year-olds who had already used generative AI in 2025 reached 44.7% in the Netherlands, 37.88% in Spain, 37.46% in France and 32.25% in Germany, according to Our World in Data based on Eurostat. Our data has no matching series for the US or China, so a direct comparison isn’t possible here.
The public response: rules and computing power
The European Union first chose to act through regulation. The EU AI Act governs uses according to their level of risk, alongside the GDPR for personal data. Our summary of what the AI Act means for you covers the details.
It is also betting on computing power. According to the European Commission’s “AI Factories” page (digital-strategy.ec.europa.eu, updated on 20 October 2026, accessed on 22 October 2026), 19 “AI Factories” and 13 Antennas are being set up around the EuroHPC programme’s supercomputers. The Commission states that EU, member state and associated country investment in these infrastructures will reach €10bn over 2021-2027. For future “AI Gigafactories” designed to train the largest models, it mentions up to €10bn in EU and national funding, expected to unlock at least €20bn in private investment.
Choosing a European AI: what changes in practice
If you’d rather use a European provider, three things matter.
- Your data. According to Mistral AI, Le Chat data is hosted in the EU by default, but training on your conversations is on by default for individual plans. Proton says it does not train models on Lumo conversations. See also our profile of Mistral’s Le Chat.
- Hosting. Open-weight models can run on your own servers in Europe. Our article open-source vs proprietary AI explains the nuances.
- The level you need. For complex coding or reasoning tasks, the top American models still lead in tests. For everyday use, the gap matters less.
| Assistant | Free | Paid from | Web search | Agent mode | Voice | Images |
|---|---|---|---|---|---|---|
| Le Chat (Mistral Vibe) | Limited | $5.99/mo | Yes | Yes | Yes | Yes |
| Lumo | Limited | $12.99/mo | Yes | — | Yes | Yes |
| ChatGPT | Limited | $8/mo | Yes | Yes | Yes | Yes |
| Gemini | Limited | €4.99/mo | Yes | — | Yes | Yes |
Frequently asked questions
Which is the leading European AI maker?
Mistral AI, founded in Paris in April 2023. It builds its own language models, releases some of them with open weights and runs the Le Chat assistant, renamed Mistral Vibe in 2026. Most other European players are service or tool companies that rely on models built elsewhere.
Does a European AI keep my data safer?
Not automatically. A European company is bound by the GDPR and often hosts data in Europe, which helps. You still need to check the settings: on Le Chat, for example, training on your conversations is switched on by default for individual plans and has to be turned off manually.
Are European models weaker than American ones?
In the ranking built from Epoch AI data, Mistral AI's best-rated model scores 79.7 out of 100 as of 22 October 2026, against 95.9 for Anthropic's two leading models. That gap covers maths, coding and agent tests. For everyday writing or summarising, the difference may be far less noticeable, and the price is often lower.
Is the AI Act holding back innovation in Europe?
Nobody can say yet with solid data. The regulation mainly places obligations on high-risk uses, and its effect on businesses will only show over several years. For a specific question about your own activity, ask a legal professional or the competent authority in your country.