Mistral Compute
Product · API / developer platform by Mistral AI
Guaranteed compute capacity on EU soil.
Mistral Compute provides guaranteed, high-performance GPU capacity on European soil for AI training and inference. It offers managed Slurm and Kubernetes orchestration on the latest NVIDIA hardware, enabling enterprises to scale sovereign AI workloads without managing physical infrastructure.
Category: AI Infrastructure
- APIs and SDKs
- Mistral API, Admin API, Regional and Priority API
- Platforms
- Mistral Cloud, Private Cloud, On-premises
- Deployment
- Cloud, European soil, Sweden site (EcoDataCenter)
- GPU Models
- NVIDIA GB200, GB300, B300
- CPU Models
- NVIDIA Grace, x86
- Networking
- InfiniBand, EVPN-VXLAN isolation
- Guaranteed GPU capacity on EU soil (Sweden site)
- Latest NVIDIA GB200, GB300, and B300 hardware
- Managed Slurm and Kubernetes orchestration
- 1 GW sovereign capacity target by 2030
Overview
Mistral Compute is a high-performance AI infrastructure platform providing guaranteed GPU capacity on European soil. It is designed for AI labs, research teams, and enterprises that need to develop and run frontier models without the operational burden of managing hardware. The platform offers access to the latest NVIDIA generation hardware, including GB200 and GB300 GPUs, delivered through managed Slurm and Kubernetes environments. Mistral Compute distinguishes itself by offering sovereign capacity at scale, with a target of 1 GW across the EU by 2030. It includes built-in enterprise controls such as SSO, SCIM, and deep observability for GPU health and performance at no additional cost.
Infrastructure and Hardware
- NVIDIA GPU Generations: Access the latest NVIDIA generation hardware including GB200, GB300, and B300 GPUs.
- CPU Options: Utilize NVIDIA Grace and x86 CPU nodes for diverse computational requirements.
- Bare-Metal Performance: Deploy bare-metal clusters on InfiniBand, instrumented by Mistral's own models for peak performance.
Control and Governance
- Built-in Observability: Access dashboards, logs, and metrics for GPU health, power, thermals, ECC, and throughput at zero extra cost.
- Enterprise Governance: Utilize SSO and SCIM out of the box with RBAC that maps directly to SLURM accounts.
Use cases
- AI labs use managed Slurm to run large-scale training and HPC workloads without operating their own clusters.
- Enterprises deploy containerized AI applications using managed Kubernetes with regional controls and enterprise identity management.
- Research teams monitor GPU health, power, and throughput using built-in observability dashboards at zero extra cost.
Specifications
- APIs and SDKs
- Mistral API, Admin API, Regional and Priority API
- Platforms
- Mistral Cloud, Private Cloud, On-premises
- Deployment
- Cloud, European soil, Sweden site (EcoDataCenter)
- Integrations
- NVIDIA GB200, GB300, B300, NVIDIA Grace, x86, InfiniBand, Slurm, Kubernetes, SSO, SCIM
- GPU Models
- NVIDIA GB200, GB300, B300
- CPU Models
- NVIDIA Grace, x86
- Networking
- InfiniBand, EVPN-VXLAN isolation
- Target Capacity
- 1 GW by 2030
- Orchestration
- Managed Slurm, Managed Kubernetes
How it works
- Capacity Reservation: Organizations contact Mistral to reserve dedicated high-performance GPU capacity in Europe.
- Environment Selection: Teams choose between managed Slurm for training or managed Kubernetes for containerized AI applications.
- Deployment and Scaling: Users deploy their models or applications using Mistral's orchestration tools, with built-in observability and governance.
FAQ
What hardware is available through Mistral Compute?
Mistral Compute provides high-performance NVIDIA GPUs, including GB200, GB300, and B300 models, along with Grace and x86 CPUs.
How does Mistral Compute handle workload orchestration?
The platform offers managed Slurm for large-scale training and managed Kubernetes for containerized workloads, allowing teams to run jobs without managing the underlying control planes.
Where is the compute capacity located?
Mistral Compute is built on European soil, with specific sites like the EcoDataCenter in Sweden, to ensure sovereign capacity and regional data control.
Does the platform include monitoring and governance tools?
Yes, it includes built-in observability for GPU health, power, and throughput at no extra cost, along with enterprise governance features like SSO, SCIM, and RBAC.
Added by AntSpace from public sources (mistral.ai).