Whose jurisdiction is your agent running in
Sovereign AI is the move to keep AI infrastructure, data and inference inside a national or regional boundary: sovereign clouds, air-gapped deployments, and nation-scale AI factories. The lens for regulated, defence and public-sector partners: where the compute physically sits and which law governs it.
| Offering | Type | Jurisdiction | Mechanism | Status | src |
|---|---|---|---|---|---|
| AWS European Sovereign Cloud Amazon Web Services |
Sovereign public cloud | EU (Brandenburg, Germany) | A structurally isolated AWS partition operated by an EU legal entity under German law: EU-resident-only operations, independent IAM, billing, DNS and certificate authority. Separate from global AWS. | Live (Jan 2026) | ↗ T1 |
| Microsoft Sovereign Cloud Microsoft |
Sovereign public + private cloud | Per-country (public + private options) | Sovereign Public Cloud and Sovereign Private Cloud tiers. M365 Copilot AI data can be processed in-country; Foundry Local runs LLM inference on NVIDIA hardware in fully disconnected environments: sovereign AI without a network connection. | GA / expanding (2025+) | ↗ T1 |
| Google Distributed Cloud (air-gapped) Google Cloud |
Sovereign / air-gapped cloud | On-prem / regional | Runs Google Cloud + Gemini models on customer-owned or in-country infrastructure, including fully air-gapped deployments with no connection to Google. Data and control plane stay inside the jurisdiction. | GA (air-gapped + connected) | ↗ T1 |
| Oracle EU Sovereign Cloud Oracle |
Sovereign public cloud | EU (operated by EU entities) | Physically and logically separate EU realm operated by EU-based Oracle entities with EU-resident support, isolated from Oracle's commercial regions. Same OCI services incl. AI/GenAI. | GA | ↗ T1 |
| NVIDIA Sovereign AI NVIDIA |
Sovereign AI infrastructure | National (multiple states) | The thesis that every nation should own AI infrastructure trained on its own data, language and culture. NVIDIA supplies the accelerators + software stack to national AI-factory buildouts. | Active (nation-scale buildouts) | ↗ T1 |
| HUMAIN AI Zone (Saudi Arabia) HUMAIN + AWS |
Sovereign AI zone | Saudi Arabia | Purpose-built in-country AI zone deploying up to 150,000 AI accelerators including NVIDIA GB300 GPUs: one of the largest sovereign AI infrastructure projects globally. | Announced / building | ↗ T2 |
Where an agent's data sits, and which legal regime governs it
Every entry here is a commercial offering, not a state programme. The operators are AWS, Microsoft, Google, Oracle, NVIDIA and, in Saudi Arabia, HUMAIN with AWS. What they sell a business is a clean answer to where its data sits and which legal regime governs it, delivered through in-region operation, structural isolation, or a fully air-gapped deployment.
Show more
The trade is coverage against control. A sovereign partition launches with a fraction of the services of the commercial region, fills in over time, and carries a price premium for the isolation. For an agent the practical consequence is jurisdictional routing: the deployment has to prove not only what it did but where it happened.
What is sovereign AI?
Keeping AI infrastructure, data and inference inside a national or regional boundary, under that jurisdiction's law. In practice that means sovereign cloud regions run by a local legal entity, air-gapped deployments, and state-backed national compute. The driver is regulation, not performance.
What is a sovereign cloud?
A partition of a cloud provider structurally separated so that only in-region staff operate it and only in-region law applies. The AWS European Sovereign Cloud, for example, runs as an EU legal entity with its own identity, billing, DNS and certificate authority. The trade-off is fewer services and a price premium.
Does it matter where an AI model runs?
For regulated work, yes. Where inference happens decides which government can compel access to the data and which rules apply to it. A healthcare, defence or public-sector deployment running on infrastructure in another jurisdiction can breach data-residency rules outright.
What is an AI factory?
State-backed compute built so a country has its own training and inference capacity rather than renting it from a foreign provider. It usually pairs domestic data centres and accelerators with a sovereign cloud, so model development and serving can happen entirely inside the border.
Is a sovereign cloud worth the cost?
It depends whether data residency is a legal requirement or a preference. Sovereign regions launch with a fraction of the services, GPU and managed-model coverage arrives later, and the isolation carries a price premium. For regulated sectors the requirement decides it. For everyone else it rarely pays.