- thekb.eu
- Knowledge graph
- Organization
- Mistral AI
Mistral AI
Mistral AI — Organization. category: French generative AI company, open and closed models, based in Paris/London, 1000 employees, valuation €12B, revenue target €1B by end of 2026, ~75% revenue in Europe · positioning: French lab that moved from open-weights model publisher to sovereign inference infrastructure provider (August 2026): announced milestones of 200 MW by end of 2027 and 1 GW by end of 2030, funded by multi-year customer commitments; hosts third-party open models, including GLM-5.2; ~$4B raised to date, Microsoft as anchor tenant of its European datacenters · role: French language model developer, "best European bet on the model layer" (SFEIR); valued at ~€20B, three years old; present in B2C (Le Chat), B2B (API), open-weights and frontier, infrastructure, verticalization · sector: AI / Open source
Arthur Mensch, questioned under oath by the Assemblée nationale's commission of inquiry on digital vulnerabilities on 13 May 2026, put the numbers on record: 1000 employees, a valuation of 12 billion euros, a target of 1 billion euros in revenue by end 2026, 1 billion euros invested in R&D, and around 75% of turnover in Europe. His framing was that artificial intelligence is a natural resource, electricity converted into tokens, and that Europe had a two year window before US hyperscalers locked up its energy capacity.
By August 2026 that framing had hardened into a business model. Timothée Lacroix described a three part announcement to VentureBeat: «Mistral Regional Endpoints» in general availability, a «Priority Tier» with committed SLAs, and multi-year commitments from «Amadeus», «ASML», «Capgemini» and «CMA CGM» converted into European Compute Units, a claim on capacity «Mistral AI» has not yet built. The announced milestones are 200 MW by end 2027 and 1 GW by end 2030. Operated capacity is under 200 MW, with three disclosed sites totalling 77 MW: 44 MW near Paris, 23 MW in Sweden with «EcoDataCenter», 10 MW at Les Ulis. Lacroix on early exit: "There is no getting out."
The same company hosts third party open models, starting with «GLM-5.2», and ships its guardrail as a separate Apache 2.0 component, «Shieldstral», whose policy is read at inference rather than baked into the weights. Two dependencies remain unresolved: «Nvidia» GPUs, and «Microsoft» as anchor tenant of the European datacenters.
- Type
- Organization
- category
- French generative AI company, open and closed models, based in Paris/London, 1000 employees, valuation €12B, revenue target €1B by end of 2026, ~75% revenue in Europe
- positioning
- French lab that moved from open-weights model publisher to sovereign inference infrastructure provider (August 2026): announced milestones of 200 MW by end of 2027 and 1 GW by end of 2030, funded by multi-year customer commitments; hosts third-party open models, including GLM-5.2; ~$4B raised to date, Microsoft as anchor tenant of its European datacenters
- role
- French language model developer, "best European bet on the model layer" (SFEIR); valued at ~€20B, three years old; present in B2C (Le Chat), B2B (API), open-weights and frontier, infrastructure, verticalization
- sector
- AI / Open source
- relations
- 42
- Cited in
- 6 fiches
Adoption measures
-
1000 employees
↻ Source ↗
-
target of €1 billion in revenue by end of 2026
↻ Source ↗
-
pricing €1 per 1M input tokens, €3 per 1M output tokens
↻ Source ↗
-
€12 billion valuation
↻ Source ↗
-
€1 billion invested in R&D in 2026
↻ Source ↗
-
internal AI budget ≈ 10% of payroll
↻ Source ↗
-
91,3 % F1 in policy adaptability, versus 61,1 % for fine-tuning on public datasets alone
-
internal productivity ×2 in 6 months
↻ Source ↗
Neighborhood
← created
← leads
→ publishes
← works at
→ collaborates with
← is based on
→ created
→ uses
← supports
→ competes with