Backfrom Nebius’s own policies · verified 2026-09-02
Nebius: data residency & retention
Headquarters
🇳🇱 Netherlands (EU)
Where inference runs
EU, US, Other
Nebius (Dutch entity) runs inference in EU datacenters (Finland, France), Israel, and the US, with each model's location shown via country flags in the console before use.
Region pinning
Yes, self-serve
Per-model datacenter location is disclosed in the console; dedicated endpoints offer contractually fixed regional deployment in EU or US datacenters with no multi-region routing.
Prompt retention
Retained · no fixed window (inference cache)
By default inputs and outputs are kept to speed up inference via speculative decoding (stored in Finland, EU); with ZDR enabled, prompts and responses are not stored after each request.
Trains on API data
No
Content is not used to train any models in either mode — retained data serves speculative decoding only.
Zero data retention
Available
ZDR is a self-serve toggle in account settings, with a warning it may impact inference speed.
Compiled from Nebius’s published privacy policy, terms, and documentation on 2026-09-02. “Not disclosed” means Nebius publishes nothing on the point. Policies change without notice, so verify against the sources above before relying on this for compliance decisions.