Mistral CEO Arthur Mensch: We’re Exploring In-House Chip Design

French artificial intelligence startup Mistral AI is actively exploring the possibility of designing and potentially manufacturing its own custom semiconductors, CEO Arthur Mensch disclosed to CNBC. This marks a significant strategic pivot for the company, signaling its ambition to gain greater control over its underlying infrastructure as it intensifies its competition with established U.S. AI giants like OpenAI and Anthropic.

“It’s certainly an area of considerable interest,” Mensch commented regarding the prospect of Mistral developing its own silicon, emphasizing that the company has not ruled out such a venture. He further elaborated that custom-designed chips could offer substantial cost reductions in deploying AI models, specifically in the efficient processing of “tokens,” which are the fundamental units of data that AI models interpret.

“The ownership of our own chips is a future consideration, one that I believe will become a necessity at some point. However, for the present, we are strategically partnered with Nvidia, who has been an exceptional collaborator. We are also actively testing various other solutions in parallel,” Mensch revealed.

Mistral AI, currently valued at nearly 12 billion euros, has established itself as a formidable player in the AI development space. Beyond its core model development, the company is making substantial investments in building out its data center capabilities, heavily leveraging Nvidia’s cutting-edge hardware. Headquartered in Paris, Mistral is widely regarded as Europe’s leading response to the dominance of American AI firms.

The company’s strategic focus is firmly on the enterprise sector, boasting prominent clients such as the global leader in chip manufacturing equipment, ASML. Should Mistral proceed with its own chip development, it would be emulating the strategic moves made by major U.S. hyperscalers like Amazon and Google, both of which have successfully designed and deployed proprietary semiconductors within their vast data center networks.

Custom silicon, often referred to as Application-Specific Integrated Circuits (ASICs), is increasingly viewed as a critical pathway for hyperscalers to exert greater control over their hardware-software ecosystems. This approach not only optimizes performance and efficiency but also allows for the creation of differentiated products and services, providing a competitive edge in the rapidly evolving AI landscape.

Mistral AI announced a significant development on Thursday: the inauguration of a new data center in France specifically engineered for AI inferencing. Inferencing is the crucial stage where AI models are actively deployed to perform tasks and generate outputs.

The company has committed an investment of 4 billion euros towards expanding its data center infrastructure across France and Sweden, aiming to dramatically increase its computational capacity. “Europe is currently at a deficit when it comes to the build-out of critical digital infrastructure, and our investments are squarely aimed at bridging that gap,” Mensch stated.

He further highlighted that Europe faces not only a technological challenge but also a significant macroeconomic one. The European bloc is increasingly recognizing AI as a strategic national asset, drawing parallels to how it has historically prioritized energy resources like natural gas. “A persistent trillion-euro commercial deficit is unsustainable if a region aims to remain competitive on the global stage. This is a realization that is gaining traction, and it should be a concern for everyone,” Mensch added.

The newly expanded compute capacity in France will serve a dual purpose: supporting Mistral’s existing clientele and providing essential resources to other AI research laboratories. While specific partnerships were not disclosed, Mensch confirmed that there is substantial demand for computational power from various AI entities. “AI labs are experiencing a severe shortage of compute resources, and we are positioned to supply some of that critical need. Many are actively seeking significant allocations of compute power from us right now,” he explained.

Mensch emphasized the need for Mistral to carefully “prioritize access” to its compute resources, ensuring that a portion is allocated to AI labs while a larger, more critical share is reserved for its enterprise customers.

In parallel with its infrastructure expansion, Mistral AI also unveiled a novel agentic platform specifically designed for enterprise clients on Thursday. This strategic move positions Mistral to directly challenge U.S. competitors such as Anthropic and OpenAI, both of whom have recently amplified their own offerings in the burgeoning agentic AI space.

The new enterprise agent platform, christened “Vibe,” is engineered to autonomously execute a range of complex tasks, including sophisticated document drafting and code generation. “Vibe represents the next frontier in agent platforms, designed to tackle immediate tasks by harnessing the power of frontier AI,” stated Timothée Lacroix, Mistral’s Chief Technology Officer. “Users can delegate complex assignments, setting clear objectives and then stepping away. Vibe is capable of conceptualizing, drafting, and delivering complete work products seamlessly from a single conversational thread. Vibe Code, a specialized module, can write, test, and deploy code across diverse codebases.”

These recent product and infrastructure announcements underscore Mistral AI’s aggressive push towards accelerated revenue growth. The company has set an ambitious target of achieving 1 billion euros in revenue by 2026. While this represents a significant leap from the 200 million euros generated in the preceding year, it still significantly trails the financial scale of its primary U.S. rivals. For context, OpenAI reported an annualized recurring revenue of $20 billion in 2025, and Anthropic is projected to generate $10.9 billion in revenue in the second quarter of 2026.