Microsoft plans disconnected AI on customer-run Azure Local

Microsoft announced on October 20 that Foundry Local, its enterprise AI service for customer-operated Azure Local infrastructure, is planned to support disconnected deployments. Running it without an internet connection requires specific platform versions, suitable worker capacity and model files staged inside the local environment. Microsoft says general availability is planned for its November 17–20, 2026, Ignite event.

What Microsoft announced

Microsoft describes Sovereign Cloud as a range of options spanning public cloud, customer-operated Azure Local and partner-operated services. The October announcement adds controls for certain engineer access to production systems in the EU and European Free Trade Association (EFTA), alongside new Azure Local capabilities.

Data Guardian is rolling out for EU and EFTA customers. It adds oversight by a trusted European partner when Microsoft engineers make qualifying remote accesses to in-scope production systems. Production Touch Ledger, also rolling out in phases in the EU and EFTA, records access and actions in tamper-evident form.

Microsoft also says external key management with Azure Managed HSM is generally available. Customers can keep the root encryption key in hardware they own and operate, or with a trusted third party, outside Microsoft infrastructure.

The offline AI announcement is for Azure Local

The announced Foundry Local deployment runs on Azure Local in customer-operated environments. Microsoft says it is intended for connected and disconnected use, with local AI inference, a broader model catalog and natural-language retrieval over local enterprise data. The catalog is described as including Microsoft, partner, open-weight and open-source models.

This is distinct from Microsoft's separate Foundry Local client app for end-user hardware. The app is another offline-capable option; the announcement's enterprise deployment is Foundry Local on Azure Local.

Microsoft announced general availability for Foundry Local on Azure Local at Ignite. Microsoft Learn describes the Azure Local service as a preview, while Ignite and that availability milestone are still in the future as of October 21, 2026.

Platform and worker requirements

For a disconnected deployment, Microsoft lists Azure Local Disconnected Operations version 2604.3.0 or later and an AKS Arc cluster running Kubernetes 1.29 or later. Its minimum and recommended worker profiles are:

These are Microsoft’s documented starting points; the capacity needed depends on the models and workload. The Azure Local requirements describe the platform and worker prerequisites.

Adding models in a disconnected environment

Operators prepare and transfer the Foundry Local extension pack and the required model expansion packs, then import them into the local edgeartifacts registry. After installing a model pack, they run POST /api/v1/models/sync; the model appears in the catalog, which operators can check with GET /api/v1/models. Microsoft’s model catalog instructions cover that update sequence.

In disconnected operations, Microsoft says authentication integrates with the environment’s Active Directory and telemetry is not transmitted to Microsoft.

Ignite milestones remain ahead

Microsoft Ignite is scheduled for November 17–20, 2026, in San Francisco and online. Microsoft also announced a public preview of Sovereign Control Panel at the event, alongside general availability for Foundry Local and GitHub Enterprise Local. GitHub Enterprise Local is described as a validated GitHub Enterprise Server appliance on Azure Local, where developers can use GitHub Copilot with locally hosted models.