In Reuters yesterday: ASML is to invest €1.3B in Mistral AI, the EU’s most prominent AI company. That’s a good thing. Investing in Mistral will bring the EU closer to being sovereign in AI.

Why is that important? The AI market is dominated by US companies, with OpenAI as forerunner. Their product ChatGPT has a huge market share.

But if your company is in the EU, you need to be sure that the data stays within the EU. The USA is not good enough.

Mistral AI is then a very capable alternative. EU based and run, it operates only its own AI models, and those are open source to boot. It’s chatbot, “Le Chat” borders on the offerings of its big American competitors.

But the recent investment, while welcome, does little to bring Mistral AI up to par with its US cousins.

📊 OpenAI, best known for its product ChatGPT, but also for DALL-E. Has a valuation of $500B. Do the math.

📊 Anthropic: it’s model Claude. For us in the software industry best known for it’s use as an AI Agent generating code. It’s market valuation sits at around $183B.

Other metrics: OpenAi has about 200 million daily active users. No such numbers from Mistral, but OpenAI has about 4 million enterprise users, Mistral just hundreds. In revenue, OpenAI dwarfs Mistral with approx. $16B vs just $60M.

For us in the EU though, Mistral is important. It is the most prominent company developing AI models.

➡️ Mistral Large is the most advanced model, multilingual (of course, it’s the EU), optimised for complex reasoning.

➡️ Codestral: understands and generates code. Not up there with the big boys yet though, with max 64k context window no match for Claude’s 1 million.

➡️ Mixtral: a cost-effective, “mixture-of-experts” set of models, best for chatbots and automation.

We are a small software company. We build special-purpose web applications, in assignment. But we are now launching our own products, of which our chatbot is the first.

So for us, the most interesting one is Mixtral. That’s because our offering now uses Mistral over API. And using an API can lead to unpredictable pricing, because, desite precautions, users might cause usage to spike.

And what would be even better than using a Mistral API for our chatbot? Running our own models.

We can do that with Mixtral. It’s model 8x7B has lower memory and compute requirements, while being more or less as capable as larger models. We could run it on an affordable single-GPU server in a colocation data centre. That would enable us to have full control over the costs, which is important for us. Predicable pricing.

We in the EU have been conveniently sitting in the shadow of US giants for 80 years. In many areas. And so also for AI.

The investment in Mistral AI by such a prominent world-wide player as ASML is step towards becoming more sovereign. Out of the shadow, into the light.