French startup Mistral AI is considering designing its own chip and may eventually develop one, CEO Arthur Mensch told CNBC.
This is Mensch’s first comment on Mistral’s semiconductor ambitions, and highlights how the company is seeking to control its infrastructure as it competes with U.S. powerhouses OpenAI and Anthropic.
“Of course it’s interesting,” Mensch said of the possibility of Mistral developing its own chip, adding that the company wouldn’t rule out that possibility.
Mensch said custom chips allow companies to “significantly reduce the cost of introducing tokens.” A token is a unit of data processed by an AI model.
“Maybe one day we will own the chip, and I think we should someday, but for now we are Nvidiais a great partner for us and we’re testing a few things here and there,” Mensch told CNBC.
Mistral, valued at about 12 billion euros, is developing AI models but is also investing in building data centers powered by Nvidia chips. The Paris-headquartered company is often seen as Europe’s answer to OpenAI and Anthropic.
The company is enterprise-focused and counts companies such as chip equipment giant ASML Among our top customers.
If Mistral develops its own chip, it would follow in the footsteps of major U.S. hyperscalers such as: Amazon and googledesigns its own semiconductors and deploys them in data centers.
Custom chips, also known as application-specific integrated circuits, are seen as a way for hyperscalers to gain more control over hardware and software integration, potentially allowing them to offer products that differentiate themselves from competitors.
Data center expansion
Mistral on Thursday unveiled a new data center in France designed specifically for inference, the actual process of running AI models.
Mistral is investing 4 billion euros in data centers in France and Sweden to expand its computing capacity.
“Europe is lagging behind in infrastructure, and we’re investing to fill that gap,” Mensch told CNBC.
He said the European Union faces not only technical issues, but also macroeconomic ones, adding that Europe is now starting to see AI as a strategic asset in the same way it looked at gas.
“We can’t afford to have a trillion commercial deficit if we actually want to stay competitive in the race. So I think it’s people realizing that we’re talking about something that should be of concern to everyone.”
The additional computing power in France will be used to service Mistral’s customers and other AI labs, Mensch said, without citing specific companies.
“AI labs have a huge need for computing, and we have some of that, but some of that actually requires us to do a lot of computing today,” Mensch said.
Mistral’s CEO said the company needs to “prioritize access” to computing, some of which will go to the AI lab, and “more importantly” access to customers.
agent focus
Mistral also announced Thursday a new agent platform for enterprises. This is aimed at competing with U.S. rivals such as Anthropic and OpenAI, which have recently ramped up their own services in this space.
The new enterprise agent platform, which Mistral calls “Vibe,” can perform tasks such as drafting and coding, the company said.
AI companies are increasingly turning to agent AI, systems that can autonomously perform tasks on behalf of users.
“Vibe is an agent platform for the task at hand, allowing us to leverage frontier AI,” Timothy LaCroix, Mistral’s chief technology officer, said in a statement.
“Users can set an outline and move forward, and Vibe will think, draft, and deliver the finished work from a single conversation. Vibe Code will write, test, and deploy code across the codebase.”
Mistral’s latest release highlights its efforts to accelerate revenue growth. The company is targeting revenue of 1 billion euros in 2026. While this figure is up from €200 million a year ago, it dwarfs OpenAI and Anthropic’s balance sheets.
OpenAI’s annual recurring revenue will reach $20 billion in 2025, and Anthropic is on track to reach $10.9 billion in revenue in the second quarter of 2026.
