Mistral AI launches cloud platform with 18K Nvidia chips

Mistral AI, the French artificial intelligence startup, has announced a sweeping expansion into AI infrastructure with Mistral Compute, a comprehensive platform built in partnership with Nvidia to compete directly with Amazon Web Services, Microsoft Azure, and Google Cloud. The Paris-based company simultaneously unveiled new reasoning models called Magistral that rival OpenAI’s most advanced systems, marking a strategic shift from purely developing AI models to controlling the entire technology stack while positioning itself as Europe’s answer to American cloud computing dominance.

The big picture: Mistral’s dual announcement signals Europe’s most ambitious challenge yet to American AI infrastructure dominance, combining sovereign cloud computing with breakthrough reasoning technology.

• The move addresses growing European concerns about dependence on U.S. technology companies for critical AI infrastructure, with the EU committing €20 billion to building AI “gigafactories” across the continent.
• “This move into AI infrastructure marks a transformative step for Mistral AI, as it allows us to address a critical vertical of the AI value chain,” said Arthur Mensch, CEO and co-founder of Mistral AI.

Key infrastructure details: Mistral Compute will run on 18,000 of Nvidia’s newest Grace Blackwell chips, initially housed in Essonne, France, with plans for European expansion.

• Nvidia CEO Jensen Huang described the partnership as crucial for European technological independence, stating “Every country should build AI for their own nation, in their nation.”
• Huang projected that Europe’s AI computing capacity would increase tenfold over the next two years, with more than 20 “AI factories” planned across the continent.
• Several facilities will have more than a gigawatt of capacity, potentially ranking among the world’s largest data centers.

Breakthrough reasoning models: Mistral unveiled its Magistral series—AI systems capable of step-by-step logical thinking similar to OpenAI’s o1 model and China’s DeepSeek R1, but with unique advantages.

• Unlike competitors that hide their reasoning processes, Mistral’s models display their full chain of thought to users in their native language rather than defaulting to English.
• The company released Magistral Small, a 24-billion parameter open-source model, and Magistral Medium, a more powerful proprietary system available through Mistral’s API.
• Early testing suggests the models deliver competitive performance while addressing speed concerns—responding to complex queries in seconds rather than minutes.

Unexpected AI capabilities emerged during training: The models demonstrated surprising abilities that weren’t specifically programmed, including multimodal reasoning and sophisticated function-calling.

• Magistral Medium retained the capacity to analyze images even though training focused solely on text-based mathematical and coding problems.
• The models gained automatic multi-step internet search and code execution capabilities naturally, without specific training for these functions.
• “Something we realized, not exactly by mistake, but something we absolutely did not expect, is that if at the end of the reinforcement learning training, you plug back the initial vision encoder, then you suddenly, kind of out of nowhere, see the model being able to do reasoning over images,” explained Guillaume Lample, Mistral’s chief scientist.

Engineering breakthrough accelerates training: Mistral developed a breakthrough in “online reinforcement learning” that allows AI models to continuously improve while generating responses.

• The key innovation involved synchronizing model updates across hundreds of GPUs in real-time, updating model weights across different GPU clusters within seconds rather than hours.
• “There is no like open source infrastructure that will do this properly,” Lample noted, describing their system as much faster and cheaper than traditional pre-training methods.
• The training process took less than one week compared to the weeks or months typically required for pre-training.

In plain English: Traditional AI training is like teaching a class where students learn from textbooks for months before taking a test. Mistral’s new approach is more like a continuous tutoring session where the AI learns and improves in real-time as it answers questions, making the whole process much faster and more efficient.

Addressing sovereignty and environmental concerns: Mistral Compute ensures European customers can keep their data within EU borders while minimizing environmental impact.

• The platform addresses two major AI development concerns: data sovereignty and carbon footprint reduction.
• Mistral has partnered with France’s national agency for ecological transition and Carbone 4, a leading climate consultancy, to assess and minimize the carbon footprint of its AI models throughout their lifecycle.
• The company plans to power its data centers with decarbonized energy sources, stating “By choosing Europe for the location of our sites, we give ourselves the ability to benefit from largely decarbonized energy sources.”

What they’re saying: Industry leaders emphasize the strategic importance of regional AI development and European technological independence.

• “We did everything from scratch, basically because we wanted to learn the expertise we have, like, flexibility in what we do,” Lample told VentureBeat in an exclusive interview.
• “Europe urgently needs to scale up its AI infrastructure if it wants to stay competitive globally,” Huang observed, echoing concerns voiced by European policymakers.
• “I think when I look at the progress internally, and I think on some benchmarks, the model was getting a plus 5% accuracy every week for like, maybe like, six weeks in all,” Lample said about the rapid improvement in reasoning capabilities.

Competitive implications: Mistral’s vertically integrated approach puts it in direct competition with technology giants while offering a complete solution from hardware infrastructure to AI models.

• The company now offers Mistral AI Studio for developers, Le Chat for enterprise productivity, and Mistral Code for programming assistance alongside its new infrastructure platform.
• With backing from Microsoft and other investors, Mistral has raised over $1 billion and continues seeking additional funding to support its expanded scope.
• The success of this European challenge may ultimately depend on whether customers value sovereignty and sustainability enough to switch from established American providers.