Analysis

Microsoft's sovereign cloud update: AI on servers you run, and oversight of who reaches your systems

1AYM is an OpenAI Select Partner.

Published . Vendor facts checked on against each vendor’s own pages, listed at the end. Plans, prices and availability change. Neither OpenAI nor Anthropic has reviewed this analysis.

The short version

Two sets of changes: one for Microsoft's public cloud, and one for the private cloud a customer runs itself. Each has its own section, with what to check before planning around it.

Public cloud controls
EU and EFTA customers get oversight and a record of Microsoft engineers' access. Customer-held root keys are generally available, and a Sovereign Control Panel enters preview at Microsoft Ignite.
Local AI in the private cloud
Foundry Local and GitHub Enterprise Local are due to become generally available at Microsoft Ignite on Azure Local. GitHub Copilot can use locally hosted models, and Microsoft 365 Local is in preview.

Public cloud: who can reach your systems, and who holds the key

What was announced

Microsoft published its update on 20 October 2026, written by Douglas Phillips, President and Chief Technology Officer of Microsoft Specialized Clouds. Microsoft introduced Microsoft Sovereign Cloud last year, and the post describes it as a range of control in three parts. Those parts are Sovereign Public Cloud, Sovereign Private Cloud and the Sovereign Partner Ecosystem. Customers choose the control for each workload and can combine the three.

Microsoft says a public-facing digital service may suit the public cloud. A system that supports essential operations may need to keep running locally if the connection fails. An AI workload that handles highly sensitive data may need its models, data and processing to stay in an environment the customer controls.

Data Guardian is being rolled out for customers in the European Union (EU) and the European Free Trade Association (EFTA). A trusted EU partner, based in Europe, oversees qualifying remote access by Microsoft engineers to in-scope production systems.

Production Touch Ledger has launched with a phased rollout in the same regions. It keeps a tamper-evident record of that access and of what was done. A tamper-evident record shows if anyone has altered it. Just-in-time Access and Customer Lockbox, which already exist, give customers more control over access to their environments.

External key management with Azure Managed HSM is now generally available, Microsoft's term for a full release. An HSM, or hardware security module, is a dedicated device that stores encryption keys.

With this option the customer keeps the root encryption key in hardware it owns and runs. That hardware can sit on the customer's own premises or with a trusted third party. Azure Key Vault Managed HSM already gave dedicated, single-tenant key protection inside Microsoft's infrastructure.

The Sovereign Control Panel enters public preview at Microsoft Ignite. It brings sovereignty settings into one view in the Azure portal. There a team can see where its sovereignty settings stand, apply and manage controls, and find gaps and risks. It also shows how workloads depend on each other and how easily they could move to another environment.

Azure Confidential Clean Rooms is in limited preview. It applies confidential computing to approved work between several organisations on shared sensitive data. Confidential computing protects data in hardware while it is being processed. Microsoft is also publishing new business continuity guidance and architecture patterns.

What it means for enterprises

A risk committee at a regulated buyer can ask who at the cloud provider is able to reach its production systems. Where Data Guardian and Production Touch Ledger apply, they give the committee a control and a record to point to.

The post names EU and EFTA customers for both. It does not mention the UK. A UK buyer who wants this oversight should ask Microsoft whether and when it applies to their tenancy, and get the answer in writing.

Holding the root key is a decision for procurement and for the security architects. The key then sits in hardware the customer controls, and that hardware and the people who run it join the customer's own estate. A buyer that already runs HSMs for other systems likely has less to set up.

Before you switch it on

Workload list
Sort your AI workloads by how sensitive their data is before choosing a control. The update is built around picking a control per workload, so the list comes first.
Coverage outside the EU
If you are in the UK or the Gulf, ask Microsoft whether Data Guardian and Production Touch Ledger cover you. The post names EU and EFTA customers only.
The root key
Decide who would own and operate the HSM, on site or through a third party, before you adopt external key management. Write down who can reach it and how.
Preview status
Treat the Sovereign Control Panel and Confidential Clean Rooms as pilots. One enters public preview at Microsoft Ignite and the other is in limited preview.

Sources: [1]

Private cloud: AI models and Copilot on Azure Local

What was announced

Sovereign Private Cloud is built on Azure Local, which the customer operates and can run with or without a link to Microsoft's cloud. Microsoft says Foundry Local will become generally available at Ignite for connected and disconnected operation on Azure Local. It comes with a larger model catalogue of Microsoft, partner, open-weight and open-source models. An open-weight model is one whose maker publishes the trained model for others to download and run.

Microsoft also says people will be able to ask questions of local company data in plain language and get answers drawn from that data. It calls this local inference and agentic retrieval, and the post gives no date for it. Microsoft 365 Local is in preview for approved customers. Microsoft says its integration brings those answers to SharePoint and Exchange servers the customer runs, and gives no date for that either.

GitHub Enterprise Local will become generally available at Ignite. It is a GitHub Enterprise Server appliance validated for Azure Local, so repositories, code and development artifacts stay under local control. Developers can also use GitHub Copilot with locally hosted models.

Confidential virtual machines on Azure Local are generally available. They encrypt memory in hardware, isolate workloads in hardware and check a workload before sensitive data or keys are released to it.

SQL Server on Azure Local is already generally available for connected and disconnected use, and DocumentDB and Blob Storage on Azure Local enter public preview. The ability to move workloads from connected to disconnected operation also enters public preview.

Microsoft says Azure Local will support multitenancy, with no date given. Several tenants would share the same hardware, each with its own resources, governance and management. Microsoft presents this as a way for local operators and service providers to offer sovereign cloud services on Azure Local.

What it means for enterprises

Microsoft's private cloud gets a model catalogue and a coding assistant on local models, both due at Ignite. Answers from local SharePoint and Exchange are announced with no date. A workload whose data rules out the public cloud can then have an AI option on the customer's own servers.

Choosing a model to run locally comes with three conditions. The model has to be in the catalogue, the hardware you run has to carry the load, and the results need their own evaluation. Check which models are in Foundry Local's catalogue before committing a use case to it.

The AI items are in preview, due at Ignite or undated, and none has a price yet. Plan pilots now, and commit to production once general availability and pricing are published.

The post does not name the Gulf. For Gulf buyers, the item to look at first is disconnected operation on Azure Local. Ask who the local partner would be and how support works. Once Microsoft gives a date, multitenancy could let a local service provider run Azure Local for buyers who would rather not run the hardware.

Before you switch it on

Model catalogue
Ask Microsoft which models Foundry Local will offer, since the post does not list them. Test the ones you would use on your own tasks.
Copilot features
Ask which GitHub Copilot features work with locally hosted models before you plan a rollout on GitHub Enterprise Local, because the post does not say.
Capacity
A local model runs on servers you pay for and run, so size them for running the model and for storage. Its speed and cost depend on that hardware.
Running offline
List the systems that must keep running if the connection fails, and test the connected to disconnected switch while it is in preview.

Sources: [1]

Where the market is heading: sovereignty covers access, keys and AI

Microsoft's update deals with four sovereignty questions for a cloud buyer. They are where a system runs, who can operate it, who holds the keys and how a service keeps running when conditions change.

The post closes by saying AI systems and agents work with sensitive data, business processes and essential infrastructure. In Microsoft's view, organisations therefore need control over where systems run, who can operate them, what they can do and how critical services continue.

Foundry Local, GitHub Enterprise Local and Microsoft 365 Local put models, a coding assistant and, later, answers from local data on infrastructure the customer operates. A buyer can then apply the same sovereignty rules to AI workloads as to its other systems. For regulated and public-sector buyers, the likely decision is which control each workload needs.

Microsoft also names National Partner Clouds such as Bleu Cloud in France and Delos Cloud in Germany. Partners run these independently, for customers who need a different ownership and operating model.

Sources: [1]

What is not known yet

We read Microsoft's post on 21 October 2026, the day after it was published, and it does not answer the points below.

Price
The post gives no price or licensing terms for any item.
Ignite timing
Several items are due at Microsoft Ignite, and the post gives no date for the event.
The UK and the Gulf
Data Guardian and Production Touch Ledger are described for EU and EFTA customers. The post does not mention the UK, the Gulf or any other country for them.
Models and Copilot
The post does not list the models in Foundry Local's catalogue, or say which GitHub Copilot features work with locally hosted models.
Dates
No date is given for local inference and agentic retrieval, for multitenancy, or for general availability of DocumentDB, Blob Storage or Microsoft 365 Local.
AI in the public cloud
The post makes no explicit data residency commitment for cloud-hosted Microsoft 365 Copilot or Azure AI Foundry. Its AI commitments are all for local use.

Sources: [1]

For engineers: keys, local models and testing disconnected operation

External keys with Azure Managed HSM

External key management is now generally available. It keeps the root encryption key in hardware the customer owns and operates, on premises or through a trusted third party. Azure Key Vault Managed HSM already gave dedicated, single-tenant key protection inside Microsoft's infrastructure.

The design has to say where the HSM sits, how Azure reaches it and who operates it. It also has to say what happens to dependent workloads if the HSM cannot be reached, and that case needs a test before production.

Sources: [1]

AI on Azure Local

Foundry Local is due to become generally available at Ignite for connected and disconnected operation. Its catalogue holds Microsoft, partner, open-weight and open-source models. GitHub Enterprise Local is a validated GitHub Enterprise Server appliance on Azure Local, and GitHub Copilot can use locally hosted models for inference.

Build an evaluation set from your own tasks and run it on the local model and on the cloud model you use today. Measure response time and throughput on your own hardware, because both depend on it.

For sensitive inference, confidential virtual machines on Azure Local are generally available. They encrypt memory in hardware, isolate the workload in hardware and attest it before sensitive data or keys are released.

Sources: [1]

Data services and disconnected operation

SQL Server on Azure Local is generally available for connected and disconnected deployments. DocumentDB and Blob Storage on Azure Local enter public preview. Moving workloads from connected to disconnected operation also enters public preview, alongside new business continuity guidance and architecture patterns.

Test the switch from connected to disconnected on a non-production workload, and record what stops working while the link is down. Multitenancy has no date yet, so plan for one tenant per Azure Local deployment for now.

Sources: [1]

Where 1AYM fits

Deciding which control each AI workload needs is part of our AI governance implementation work, sold as our Enterprise AI Platform & Agentic Workflows engagement. We build identity and permissions the system enforces, and an audit trail produced as the controls run. Where a region is a requirement, data residency is architected in, and we can say which region each store sits in.

We choose models on the evidence from each workload, across OpenAI, Anthropic, Gemini and open-source. On a private cloud, that evidence decides which model from a local catalogue runs on your own servers. We work inside our clients' ISO and SOC 2 compliant environments, including banks and other highly regulated organisations.

To find which workloads are sensitive enough to need a private cloud at all, we would start with our AI Opportunity & Feasibility Sprint. It is fixed scope and typically runs two to four weeks.

Sources

  1. [1]Microsoft, Advancing digital sovereignty with greater control and choice, Douglas Phillips, 20 October 2026 (checked 21 October 2026)

Frequently asked questions

What did Microsoft announce on 20 October 2026?

An update to Microsoft Sovereign Cloud. In the public cloud, it adds oversight and a tamper-evident record of Microsoft engineers' access for EU and EFTA customers, and customer-held root keys. In the private cloud on Azure Local, Foundry Local and GitHub Enterprise Local, with GitHub Copilot on local models, are due at Microsoft Ignite. Microsoft 365 Local is in preview.

Can Microsoft's AI tools run without a connection to its cloud?

Microsoft says Foundry Local will become generally available at Ignite for connected and disconnected operation on Azure Local, which the customer runs. GitHub Enterprise Local, also due at Ignite, will let developers use GitHub Copilot with locally hosted models.

Do Data Guardian and Production Touch Ledger apply in the UK?

The post describes both for customers in the EU and EFTA, and does not mention the UK. A UK buyer should ask Microsoft whether and when they apply.

What is external key management with Azure Managed HSM?

It is an option, now generally available, for keeping the root encryption key in a hardware security module the customer owns and operates. The module can sit on the customer's own premises or with a trusted third party.

How much does it cost?

Microsoft's post gives no price or licensing terms for any of these items.

Further

Sensitive AI workloads with nowhere to run?

Half an hour is enough to sort your AI workloads by sensitivity and agree which need a private cloud or a key you hold.

Last reviewed · 1AYM