Dissecting the $6 billion valuation myth
Mistral AI reached a $6 billion valuation in June 2024 through a Series B round that raised €600 million. This figure is outdated. In September 2026, the company completed a €3 billion Series D round, which resulted in a post-money valuation exceeding €21 billion and saw Samsung Electronics leading alongside co-leads Scaleup Europe Fund and PSG Equity, which is managed by EQT. This capital expansion supports frontier research and the development of infrastructure to scale compute capacity. Mistral targets annual recurring revenue exceeding $1 billion before the end of 2026. The company reported revenue of approximately $400 million in February 2026, an increase from the $30 million it earned in 2024. In September 2025, ASML led a €1.7 billion Series C round that valued the company at €11.7 billion. This funding followed a Series A in December 2023 that raised $415 million and established a $2 billion valuation. The early seed round in June 2023 raised €105 million and valued the company at €240 million. Lightspeed Venture Partners led that initial round, which included investors like Eric Schmidt and Xavier Niel. The Series B round, led by General Catalyst and DST Global, included participation from Salesforce Ventures, IBM, and Cisco.
The reality of Llama 4 performance and benchmarking
A Meta executive denied claims that the company artificially inflated Llama 4 benchmark scores to deceive evaluators. This controversy surrounds the integrity of AI performance metrics in the frontier model race. Mistral uses a different approach, relying on sparse Mixture-of-Experts architectures and memory optimization to reduce inference costs. For developers, the Codestral model provides 22 billion parameters and supports 32,000 token prompts. This exceeds the prompt capacity of Llama 3 70B. Mistral also provides the Mistral Large 3 model, which has a standard API rate of $0.50 per million input tokens and $1.50 per million output tokens.
| Model | Type | Parameter/Feature | API Pricing (per 1M tokens) |
|---|---|---|---|
| Mistral Large 3 | Proprietary/Open-weight | Multilingual | $0.50 Input / $1.50 Output |
| Codestral | Open-source coding | 22B parameters | Variable |
| Llama 3 70B | Open-source | 70B parameters | Variable |
The Codestral model understands more than 80 programming languages. You should ignore the idea that Mistral only provides API access. The company also sells customized deployments and local hosting. The Vibe platform, formerly known as Le Chat, functions as a unified agent for productivity and coding. Users find that Mistral Medium 3 outperforms several competitors in certain tasks.
Enterprise adoption and infrastructure control
Mistral supports over 125 global enterprises across 20 countries, including Airbus, ASML, and HSBC. These companies use Mistral for mission-critical tasks in manufacturing, finance, and defense. Clients include BMW, Stellantis, CMA CGM, and Helsing in addition to Airbus and HSBC. The company provides a full stack including open-weight models, infrastructure, and compute capacity. This allows organizations to maintain control over data and deployment. Mistral also developed the Vibe platform to serve as a unified agent for productivity and coding. The company expanded its Microsoft partnership in July 2026, which involves a multibillion-dollar commitment to fund data center growth in Europe. Microsoft will use capacity from Mistral’s Europe-based data centers to serve cloud customers. Mistral also plans to build a 23 megawatt site in Sweden to generate revenue.
The Microsoft deal includes the availability of Mistral Medium 3.5 and OCR 4 in Microsoft Foundry. Mistral Medium 3.5 is also available in Microsoft Copilot Studio. Mistral uses NVIDIA Rubin and Vera Rubin platforms to scale its infrastructure. The company’s compute capacity relies on hardware like the NVIDIA H100, H200, and L40s. The founders, Arthur Mensch, Timothee Lacroix, and Guillaume Lample, previously worked at Google DeepMind and Meta. They established the company in April 2023. Mistral also announced a tie-up with Cloudera in September 2026 to bring specialized intelligence to enterprise data workloads. The startup collaborated with TotalEnergies in September 2026, investing more than €100 million in advanced AI models for oil-and-gas applications. Mistral hired Marc Oman into its compute infrastructure unit to manage energy and power strategy. Does the heavy reliance on NVIDIA hardware prevent companies from achieving true technological sovereignty?
