Skip to content

Sovereign AI

Residency in the contract, not in a FAQ.

For workloads that cannot leave the country, the useful question is not whether a provider says it keeps data in-region. It is whether the platform can show you, per model and per key, and stop the call when the answer is no.

  1. Residency is recorded, not inferred

    Each model carries an explicit in-region flag and a named jurisdiction, set by an operator. A badge is never derived from a provider's name or a model's prefix, because those are guesses and a guess is what a compliance review is trying to eliminate.

  2. Keys can be confined to in-region models

    An API key carries an allowlist. Scope one to in-region models only and a service holding it cannot send a regulated payload to a model outside the country, whatever the calling code does.

  3. The jurisdiction is the operating one

    The campus sits in Batam under Indonesian jurisdiction, operated by a Singapore-incorporated entity. Residency follows where the workload runs, not where the company is registered.

  4. Every mutation is on the record

    Registering a cluster, minting a key, changing a rate, deploying a workload — each writes an audit entry, including the attempts that failed. That log is what an auditor reads.

The honest part

The campus is still in development.

The residency machinery above is built and running. The Batam capacity it is designed to serve is not finished, and until it is, an in-region model means one served from a provider we have verified sits inside Indonesia. If your timeline depends on ours, ask — we will tell you what is committed and what is a plan.

Ask about timing