Sovereign AI readiness check
01 Document your data rights and owners
List the datasets the AI would use, who owns each and for which purposes you may use it. Without that, a self-hosted model does not solve the legal question.
02 Map residency requirements to hosting options
Write down where each dataset may be processed, then compare Nebius in a European region, your own servers and Azure in an EU region against that list.
03 Build an evaluation set before you pick a model
Collect real examples with the answers you expect and an acceptance threshold. Then compare hosted and open-weight models on evidence, not on a demo.
04 Prepare a private foundation
Private endpoints, Entra ID groups, Key Vault and monitoring should exist before the first model is deployed. They are the same whether you host on Azure or on your own servers.
05 Choose the hosting and run a self-hosted pilot
With the groundwork done, pick one use case, one open-weight model and one hosting option, fine-tune on your own data where it helps and plan the handover from day one.
A checklist is a starting point. It lists what we would ask first, not everything a platform or pilot needs.