Vast Data, an AI operating system company, shared more details at the Mistral AI NOW Summit about its partnership with Mistral and Mistral Compute to power a new generation of Nvidia-accelerated AI factories in Europe.With Mistral Compute adopting the Vast AI operating system as the data foundation for large-scale Nvidia GB300 NVL72 systems, the collaboration brings together the three layers that production AI requires: accelerated compute from Nvidia, frontier models from Mistral AI, and the unified AI data platform from Vast.

The deployment brings one of Europe’s largest concentrations of Nvidia GB300 NVL72 capacity online, helping power the European and global AI economy. Mistral Compute will operate the infrastructure, with Vast providing the AI data platform services that unify data access, movement, governance and performance across the full estate.

The Vast AI operating system is purpose-built for production AI factories. At Mistral Compute, it will give researchers, applications and agents access to a common data environment across training, inference, retrieval and enterprise AI deployment – without copies, silos, or stitched-together pipelines. For Europe, where AI capacity is becoming a strategic priority, the combination of Nvidia accelerated compute, Mistral models, and the Vast data platform offers a path to building advanced AI with greater local control over where data lives, how it is governed, and how it is put to work in production.

A unified foundation for production AI
In a modern AI factory, researchers, applications and agents all need fast, consistent access to the data, context and memory their workloads depend on. Without a data layer that can keep pace, accelerated compute cannot deliver its full value or throughput economics an AI factory depends on.

For AI factories to operate in production, the underlying platform must support the full range of data and runtime state modern AI now depends on, from files and tables to vectors, KV cache, event streams and persistent agent memory, across the sites and geographies a frontier program now spans. This is what Mistral Compute is building with Vast.

The collaboration spans from research to AI cloud infrastructure.

Vast is already deployed at the leading AI clouds Mistral AI relies on for model training and research. As Mistral’s researchers move workloads across cloud environments, Vast DataSpace provides a consistent unified namespace across distributed environments so teams can access, move and manage data without re-architecting pipelines for each cloud. Mistral Compute is now operating its own Nvidia-accelerated AI cloud platform as an Nvidia Cloud Partner, having adopted the Vast AI OS as its core AI data platform. Vast is already in production supporting Mistral AI’s internal workloads as well as end-customer workloads running on Mistral Compute’s cloud services. The GB300 NVL72-based AI factory deployment extends that foundation to a new tier of scale.

As Mistral AI brings frontier models into enterprise environments, customers need a platform that can connect those models to their own data with the performance, governance, isolation and control that production AI requires. Vast provides a common AI OS foundation for these deployments, helping enterprise customers run Mistral-powered AI closer to the data their applications and agents depend on.

Together, these tracks make Vast the unifying operating foundation for Mistral’s AI lifecycle. This matters because AI sovereignty is ultimately exercised at the data layer. For enterprises and public sector organizations, control depends on where data lives, how it is governed, how it moves between training and inference, and how models are deployed vs. sensitive enterprise data. By providing a common data foundation across research, cloud, and enterprise deployment, Vast helps Mistral Compute support AI workloads that require both frontier performance and trusted control.

Quotes
“A modern AI factory is a distributed system of compute, data, memory, and models, and the architecture has to be designed as one from the ground up. That means scaling linearly with the model, working across the sites and geographies a frontier program now spans, and handling every shape of data the system depends on, from KV cache to vectors to memory to the traces that feed the next round of training. With Mistral and Nvidia, we’re moving from large GPU clusters to durable production platforms, and the Vast AI OS is the data layer underneath,” said Alon Horev, co-founder and CTO, Vast Data.

“AI Infrastructure needs to be purpose-built to serve the unique needs of large-scale training and inference. With Nvidia GB300s and the Vast Data AI platform, Mistral is building frontier infrastructure to serve researchers, enterprises, and customers globally,” said Yann Léger, VP, cloud compute, Mistral AI.

“AI factories are essential infrastructure for every industry. Together, Vast, Mistral and Nvidia are bringing sovereign AI factory capacity online in Europe, combining accelerated computing, AI models, and a unified data platform into one production-ready service – enabling organizations to build AI platforms that are ready for real-world enterprise and sovereign AI workloads and scale them with the performance and flexibility needed for long-term growth,” said Rod Evans, EMEA VP, supercomputing, AI factories and Nvidia cloud partners, Nvidia. 

Availability and what’s next
Vast is in production today across Mistral Compute, with additional capacity coming online as the GB300 NVL72-based AI factory expands. Mistral’s adoption of Vast AI OS services for model development is underway.

At the Mistral AI NOW Summit, Vast’s Alon Horev joined Mistral AI’s Yann Léger and Nvidia’s Rod Evans on a panel titled “No Shortcuts: AI Infrastructure Is the New Critical Infrastructure,” moderated by Mistral AI’s Brooke Gleason. The session explored how production AI infrastructure is being designed across compute, data, models, and applications, and what it takes to bring an AI factory online in production environments.