The deciding factors are memory per GPU, GPU-to-GPU bandwidth and how long you need the capacity. Everything runs on bare metal, in dedicated clusters, with direct hardware access.
Foundation-model training and large fine-tunes that need the whole model, optimiser state and activations close to the GPUs, across several nodes.
Serving open-source or proprietary models from dedicated European nodes, with predictable cost per GPU-hour and no shared tenancy.
Agent workflows, multi-application pipelines, simulation and professional graphics that mix AI with rendering and interactive work.
For these customers, dedicated nodes in the EU under a European operator are a requirement. The arrangements below are contractual; certification status is stated on the Security page.
Dedicated European clusters for resale or for listing on a compute marketplace, reserved on terms sized to your demand. Partner pricing is set per contract.
Talk to our team →Europe's emerging AI companies can apply for GPU capacity in exchange for an equity stake, with introductions to investors, banks and advisers.
Compute for equity →Every node is delivered as bare metal: your operating system, your drivers, your orchestration, with direct access to the GPUs. Virtual machines, Kubernetes and SLURM are in development on the same infrastructure.
Describe the model, the data and the timeline. You get a recommendation and a quote, not a sales deck.