Europe Is Backing Its Own AI Stack

Europe raised $25 billion in venture capital last quarter, its strongest quarter in four years. Three of every four dollars went to AI. The single largest cheque went to Mistral: €3 billion, the biggest venture round a European company has ever raised. On October 29, Mistral released the weights of Large 4, its new flagship model, and halved its price. Releasing the weights means anyone can download the model and run it on their own servers. Europe is no longer just funding AI. It is buying its own.

Mistral was founded in Paris in April 2023 by three French AI researchers. Arthur Mensch, the CEO, worked on large language models at Google DeepMind. Guillaume Lample and Timothée Lacroix came from Meta, where both worked on the original Llama model. The three had known each other since their student days in France.

They left Google and Meta with one thesis. Europe needed its own frontier models, and those models should be open and efficient enough for any company to run. They named the company after the mistral, the cold wind that blows through southern France.

Mistral has set European records from the start. Weeks after its founding, it raised Europe's largest ever seed round. A little over three years later, it is worth nearly 90 times its seed price.

The leads tell the second story. American venture funds led Mistral's first three rounds. The chip industry led the last two. ASML is the Dutch company that builds the only machines able to print the world's most advanced chips. It led the Series C. Samsung led the Series D. The EU's Scaleup Europe Fund and PSG Equity co-led, and Nvidia invested again.

So Europe's sovereign AI lab is financed by a Korean chipmaker, a Dutch chip equipment maker, an American GPU maker and Europe's own scaleup fund. That is not a contradiction. Sovereignty is not about who owns the shares. It is about who holds the weights and where the model runs. Mistral keeps both in Europe. The capital can come from anywhere.

At a trillion parameters, Large 4 is about one and a half times the size of Large 3, Mistral's previous flagship. Mistral trained Large 4 in about two months in its own data centers in Europe. In March, Mistral borrowed $830 million to build its own data center south of Paris. The new €3 billion pays for research and more computing power.

This is what a European buyer pays for. A bank or manufacturer can use Large 4 through Mistral's European cloud or host it itself. Either way, its data stays in Europe. The license allows free use, modification and commercial use. Before the release, Mistral tested the model with cybersecurity leaders, vetted partners and state authorities. It works in more than 160 languages, which in Europe is a requirement, not a feature. Airbus, ASML and HSBC are already Mistral customers.

Large 4 is not the best model in the world, and it is not priced like it. On a finance benchmark Mistral published at launch, it scored 54.7. That beat OpenAI's GPT-6 Astra at 53.5 and trailed Anthropic's Claude Opus 5.5 at 58.6. Independent testing of the released model is more measured. It ranks 17th of 32 models overall and fifth among open weight models.

The price closes the argument. Large 4 costs a sixth to a tenth as much as Claude Opus 5.5: $0.70 and $2.10 per million input and output tokens, against $4 and $20. It is already cheaper than Google's Gemini Flash. When Google's introductory rate ends on January 1, it costs less than half as much. For a regulated buyer, control used to come at a premium. Now it comes at a discount.

Independent testing comes first. Today the strongest numbers for Large 4 come from Mistral itself. Because the weights are public, any lab or customer can now download the model and run its own tests. If those results match Mistral's, the case holds.

Then contracts. A pilot is a trial a company can walk away from. A production contract means the model runs real work every day and Mistral earns recurring revenue from it. When European banks, manufacturers and ministries name Large 4 in production, sovereignty has a paying customer.

Last, what Mistral builds next. Large 4 is the base for a new family of specialized Mistral models, and those are what a regulated buyer will actually deploy.

Investors have made their bet on a European model. Now I want to see who buys it. 

Bashar Aboudaoud
Managing Member, UpRound

Share with 10 people to receive a quarterly call with Bashar.