Notes from Mistral's AI Now Summit: A Full-Stack Bet on Europe

Mistral AI held its AI Now Summit in Paris on 29 May 2026. The developer Koen van Gilst attended and published his notes the same day. His summary: the French company no longer presents itself as a model vendor alone, but as a builder of the whole AI stack — compute, models, platforms and consultancy.

Compute, models and partnerships

Mistral owns its compute: a 40MW data centre in Paris, with more planned, including one in Sweden. It stresses efficient, open and bespoke models that customers own and can run on-premises, which the author sees as its main differentiator against Anthropic and OpenAI. The summit focused on partnerships rather than new models — ASML, BNP Paribas and Amazon's Alexa+ — framed as solving real problems. Van Gilst found the lack of technical novelty disappointing, though Mistral did launch Vibe for Work, a product in the same category as Claude for Work.

The harness matters more than the model

In a talk by Pieter Stock, the argument was that the model alone is not enough. A harness adds context, persistence and learning, while reasoning is what lets a system backtrack, recover from errors and stay transparent. Skills, developed together with the AI agent, are how organisations capture their best practices.

Small specialised models

Mistral's strategy is small, specialised models. The summit showed cases where fast, focused models beat large general-purpose ones on energy efficiency and speed: Document AI for OCR, used by the European Patent Office for large-scale document processing; Voxtral for multilingual voice, powering Alexa+ in Europe; and Robostral for industrial robotics with ASML. In token-heavy agentic applications, speed and efficiency now matter as much as raw capability.

Sovereignty, on-prem and 180,000 papyri

Sovereignty remains the core pitch. BNP Paribas runs Mistral models on-prem for KYC work in Belgium, keeping sensitive data inside the bank, and Abanca uses agent orchestration for more than a million customers in its app. For regulated European industries that is an alternative to US hyperscalers. One unusual talk covered ancient papyri: a team from the Austrian Academy of Sciences fine-tuned Mistral's Codestral to read fragments of discarded documents unpublished for decades, helping open a collection of 180,000 items found in the Egyptian desert — work that would otherwise have taken more than 2,000 years.

The summit left the impression of a company that is not chasing AGI but positioning itself as Europe's full-stack AI partner with returns today. Whether that pays off depends on European companies committing to it — though the era of blindly trusting US tech giants, the author writes, is coming to an end.