$10.7 billion, the lithography machine giant leads the investment in an AI startup, which has caught the eyes of Huang Renxun and Tim Cook. Three post-90s entrepreneurs are going all out.
On September 8, according to Reuters' report today, ASML, the Dutch semiconductor lithography giant, will participate in the Series C financing of French generative AI unicorn Mistral AI, contributing €1.3 billion (approximately $1.5 billion or RMB 10.7 billion) to become its largest shareholder and secure a seat on the board. The total Series C financing is €1.7 billion (around $2 billion or RMB 14 billion).
After this round of financing, Mistral AI's valuation is expected to reach €12 billion (about $14 billion or RMB 100.3 billion), nearly doubling from June last year.
Founded in April 2023 by three post - 1990s entrepreneurs, Mistral AI is the largest AI unicorn in Europe. It has previously completed multiple rounds of financing with participation from well - known companies and investment institutions such as former Google CEO Eric Schmidt, Databricks, NVIDIA, Microsoft, Salesforce, and a16z. The total amount of the four rounds of financing has reached approximately $1.175 billion (about RMB 8.4 billion).
According to foreign media The Information, Apple executives have internally discussed the possibility of acquiring Mistral AI.
The founders of Mistral AI are former DeepMind researcher Arthur Mensch and two former Meta AI scientists, Timothée Lacroix and Guillaume Lample.
This unicorn, focusing on open - source model research and development, has launched various types of open - source models, including voice, programming, and multi - modal models, with dozens of models covering the cloud, edge devices, and different parameter scales.
In June 2023, just two months after its establishment and before releasing its first batch of models, Mistral AI secured €105 million (about $130 million or RMB 900 million) in financing, setting a record for the largest seed - round financing in European history at that time. In December 2023, it completed a Series A financing of $415 million (about RMB 3 billion). In June 2024, it received €600 million (around $700 million or RMB 5 billion) in Series B financing, and its valuation reached €6 billion (about $7 billion or RMB 50.1 billion).
As a European technology company, this investment may deepen the cooperation between Mistral AI and ASML. ASML manufactures lithography machines for top - tier chip production and uses AI to improve the efficiency of its tools. It can leverage Mistral AI's data analysis and AI capabilities to enhance the performance of its tools and develop more products.
01. Founders from DeepMind and Meta Team Up, with Over $2.26 Billion in Financing
The three founders of Mistral AI are all post - 1990s.
Arthur Mensch was born in 1992. He graduated from top universities such as École Polytechnique, Université Paris - Saclay, and École Normale Supérieure. He obtained his Ph.D. in machine learning at the age of 26 and then conducted two years of post - doctoral research in mathematics. After graduation, he joined Google DeepMind as a research scientist, working on large - language - model - related projects. He left Google DeepMind to start his own business in May 2023.
During his tenure at Google DeepMind, he participated in the development of the Chinchilla model, which demonstrated that with sufficient computing resources and data, small models can outperform large models through long - term training.
Arthur Mensch's LinkedIn profile
Guillaume Lample and Timothée Lacroix are the main architects of Meta's Llama model. Guillaume Lample graduated from École Polytechnique and Carnegie Mellon University and has worked at Meta for over six years after graduation.
Timothée Lacroix's LinkedIn profile (left), Guillaume Lample's LinkedIn profile (right)
The three co - founders met while studying at École Polytechnique. They aimed to establish a powerful company that could make an impact. The name "Mistral AI" comes from French, meaning "a strong wind blowing from France".
Co - founders of Mistral AI: Guillaume Lample, Arthur Mensch, and Timothée Lacroix (from left to right)
Its mission is to make top - tier AI accessible and transparent, so it has focused on open - source models since its inception. Currently, Mistral AI has released over 10 large models, covering various types from the cloud to the edge and from language to multi - modal models. In March this year, Arthur Mensch mentioned in an interview with foreign media Business Insider that they compare DeepSeek to the Chinese Mistral AI, and for open - source models, (the emergence of DeepSeek) is a great moment.
In terms of financing, Mistral AI has received four rounds of financing, with a total of over $1.175 billion (about RMB 8.4 billion). If the latest financing is completed, the total financing will reach $3.175 billion (about RMB 22.6 billion).
Notably, although Mistral AI is a French startup, many of its investors are from the United States, including tech giants like Microsoft, NVIDIA, and Salesforce, as well as well - known Silicon Valley venture capital firms such as General Catalyst and a16z.
In addition to investment, in February last year, Microsoft also announced a partnership with Mistral AI. Microsoft will hold a small stake in Mistral AI, and Mistral AI's open - source and commercial large - language models will be available on Microsoft's Azure AI platform. It's worth noting that Mistral AI is the second company to offer commercial large - language models on the Azure AI platform after OpenAI.
02. Mistral AI: Open - Source Models, AI Products, and Data Centers
Since its establishment, Mistral AI has achieved many milestones in the field of open - source large models.
These include the MoE model Mixtral 8x22B, launched on April 10, 2024, with a parameter scale of 176 billion, second only to Elon Musk's Grok - 1, becoming the second - largest open - source model in terms of parameter scale at that time. In May this year, it launched Mistral Medium 3, which has performance comparable to Claude Sonnet 3.7 but a lower cost than DeepSeek - V3.
Mistral AI's models are not completely free and open - source. Its model licenses are divided into three categories: weight access under the Apache 2.0 license, Mistral AI research license, and Mistral AI commercial license. Models open - sourced under the Mistral AI research license cannot be used for commercial deployment, creating derivative products, fine - tuning, etc.
Model open - source rules
Most of the company's models are open - sourced under the Apache 2.0 license. The edge - side model Ministral 8B, the flagship large - language model Mistral Large, and the multi - modal model Pixtral Large are available under the research license.
According to its official website, the Mistral series of models are applicable to three types of tasks. Mistral Nemo can support simple batch - execution tasks such as classification, customer support, or text generation. Mistral Small supports intermediate tasks that require medium - level reasoning capabilities, such as data extraction, document summarization, email writing, job - description writing, or product - description writing. Mistral Large supports complex tasks that require strong reasoning capabilities or high specialization, such as synthetic - text generation, code generation, RAG, or Agent.
Mistral AI model series
On the product side, in February this year, Mistral AI launched its chatbot Le Chat, which competes with ChatGPT. Within less than two weeks of its launch, the mobile - app downloads exceeded one million, topping the free - download list on the French iOS App Store.
French President Emmanuel Macron even "advertised" for this product during a TV interview before the Paris AI Action Summit: "Download Le Chat developed by Mistral AI instead of ChatGPT from OpenAI or others."
In July this year, Mistral AI added features such as in - depth research, native multi - language reasoning, advanced image editing, and a project function to Le Chat. The project function allows users to group and save chats, documents, and ideas.
Le Chat feature updates
Compared with ChatGPT Enterprise or Claude Enterprise, Mistral AI's advantage is that it allows enterprises to deploy Le Chat in their own environments and supports custom models and user interfaces to meet the needs of data - sensitive enterprises.
In addition to focusing on models and products, Mistral AI is also investing in AI infrastructure. It is collaborating with NVIDIA to build a 40 - megawatt data center near Paris, powered by 18,000 NVIDIA chips.
03. Revenue in the Tens of Millions of Euros Last Year, with Multiple Corporate Partnerships
In terms of profitability, Mistral AI's revenue lags far behind that of leading large - model startups like OpenAI. In July, TechCrunch reported, citing multiple sources, that Mistral AI's revenue in 2024 was approximately tens of millions of euros. In June this year, Arthur Mensch revealed in an interview with the Financial Times that the company's revenue has tripled since the beginning of 2025.
At the end of July, according to The Information, OpenAI's annual recurring revenue (ARR) in the first seven months of 2025 reached $12 billion, far exceeding the approximately $4 billion in 2024. Its competitor Anthropic's annual recurring revenue this year reached $4 billion, nearly quadrupling from the beginning of the year.
In terms of its profit model, many of Mistral AI's products are free. It also offers free packages for developers, allowing them to use the models for fine - tuning and building test applications.
Paid products include Le Chat and API monetization.
The subscription plan for Le Chat charges individual users $14.99 per month (about RMB 107) and team users $24.99 per month (about RMB 178), with different limits on the number of generated answers.
In the pure B2B field, Mistral AI monetizes its main models through usage - based API pricing. The company likely also earns a significant portion of its revenue from its strategic partnerships.
Arthur Mensch revealed in a TF1 TV interview in February this year that Mistral AI has partnered with French labor organizations and Veolia. By the end of this year, automakers behind brands such as Peugeot, Citroën, Fiat, and Jeep will update their mobile apps, replacing the owner's manuals with AI assistants based on Mistral AI.
In addition, in April this year, Mistral AI launched a five - year, €100 - million AI strategic cooperation plan with French shipping giant CMA CGM, integrating Mistral AI's technology into CMA CGM's logistics and customer - service areas.
The new financing that Mistral AI is about to receive may help it further expand its revenue scale to match its €12 - billion (about $14 - billion or RMB 100.3 - billion) valuation.
04. Conclusion: Three - pronged Strategy to Narrow the Revenue Gap with OpenAI
Looking at Mistral AI's strategy, this startup has established three major business areas: models, products, and infrastructure. Mistral AI continuously releases open - source models, building a model matrix with different parameter scales and modalities. At the same time, it launched the chatbot Le Chat, which competes with ChatGPT, and has attracted a large number of users. This year, the startup announced its entry into the AI - infrastructure market, planning to build data centers in Europe, demonstrating its strong momentum in catching up with OpenAI in the large - model field.
However, in terms of revenue, Mistral AI still lags far behind leading large - model startups like OpenAI and Anthropic. This new round of financing may accelerate Mistral AI's infrastructure construction and global business growth, further narrowing the gap between its revenue - generating ability and its high valuation.
This article is from the WeChat official account "zhidxcom", author: Cheng Qian, editor: Xin Yuan. Republished by 36Kr with permission.