Mistral AI

Mistral AI is the Paris-headquartered frontier LLM lab founded in April 2023 by Arthur Mensch (CEO), Guillaume Lample, and Timothée Lacroix, three former DeepMind and Meta AI researchers who built the company on a strategy of open-weight foundational models plus commercial frontier offerings. Since 28 May 2026 the assistant formerly called Le Chat is Vibe: one product and one subscription with three modes, Vibe Work for delegating multi-step tasks across connected apps and tools, Vibe Chat for the turn-based conversation Le Chat was known for, and Vibe Code for developers as a command-line tool, a VS Code extension and a web interface. Accounts, plans and conversations carried over, and chat.mistral.ai still opens the service. The developer API once branded La Plateforme now sits in Studio, together with agent building, fine-tuning and batch processing; Forge covers custom model development, and Applied AI Services handle tailored R&D. Customer references include Stellantis (automotive), ASML (semiconductors), CMA CGM (logistics), Airbus and HSBC. The current open-weight line-up is released under Apache 2.0: Mistral Large 3, Mistral Small 4, the Ministral 3 family and the Voxtral transcription models, alongside the earlier Mistral 7B and Mixtral. Mistral Medium 3.5 ships under a modified MIT licence, while Codestral, the OCR models and the embedding models remain API-only. Studio also serves third-party open models, starting with Z.ai's GLM.

Funding trajectory: seed €105M (June 2023) at €240M led by Lightspeed; Series A US$415M (Dec 2023) at US$2B led by Andreessen Horowitz; Series B US$640M (June 2024) at US$6B; Series C €1.7B (September 2025) at €11.7B valuation led by Dutch chip giant ASML with a €1.3B cheque for an 11% anchor stake; March 2026 US$830M debt financing from a consortium of seven banks; and a €3B Series D (September 2026) at a post-money valuation above €21B, led by Samsung Electronics with EQT's Scaleup Europe Fund and existing investor PSG Equity as co-leads. Advent, funds managed by BlackRock and the Grand Duchy of Luxembourg joined as new investors, and returning investors include Andreessen Horowitz, ASML, Bpifrance, BNP Paribas CIB, DST Global, General Catalyst, Index Ventures, Lightspeed, NVIDIA and Salesforce Ventures. European industrial and public backers sit next to a deep bench of US venture, growth and private-equity investors, which is why ownership is recorded as EU-Sitz, US-finanziert rather than EU-owned.

Procurement-grade picture: the contracting party is Mistral AI SAS under a versioned DPA, effective 27 July 2026 and readable without login, and the Vanta-hosted trust centre now names the sub-processors behind the managed services. Inference for Vibe and Studio runs on a mix of US-owned and Mistral-owned infrastructure: Microsoft (cloud infrastructure in Sweden and Norway), Google (Netherlands and Belgium, plus the United States for the US endpoint), CoreWeave (inference in the EEA) and Mistral Compute, the company's own infrastructure subsidiaries in France. Cloudflare sits in front of every product, and processors for web search, image generation, payments and phone verification are listed with processing in the United States. The help centre states that data is hosted in the EU by default. Since Mistral Regional Endpoints went generally available on 11 August 2026, API customers can pin inference to an EU or a US endpoint at a 10% premium, while the default global endpoint does not commit to a location and operational metadata may be handled outside the chosen region. Because three US-owned clouds sit in the core inference path even in European regions, the at-rest exposure for managed-service customers is recorded as Erheblich under the strict CLOUD Act stance. Mistral holds ISO 27001, ISO 27701 and SOC 2 Type II, with reports available on request through the trust centre. It is also building its own European capacity, including a 10 MW inference site at Les Ulis (Essonne) announced in May 2026 for the third quarter, but Microsoft, Google and CoreWeave remain on the sub-processor list. The open-weight self-host path is the cleanest sovereign answer: Mistral Small 4, Ministral 3 or, on multi-GPU nodes, Mistral Large 3 running on Hetzner H100, OVHcloud GPU, Scaleway B300, STACKIT GPU, or IONOS sovereign cloud removes Mistral and its sub-processors from the data path. Best fit: French and EU regulated buyers, defence and public-sector procurement, large enterprises in automotive / semiconductors / logistics, and any organisation evaluating sovereign frontier LLMs, but procurement-led EU-only buyers should choose the self-host path on EU GPU infrastructure rather than the managed services.