Blog Post

Azure Arc Blog
2 MIN READ

Protect Your AI Models and Data at Edge: Confidential Computing on Azure Local

Dona_Mukherjee's avatar
Nov 18, 2026

AI is no longer confined to the cloud. It's running on factory floors, in hospitals, at energy sites and in sovereign datacenters, often on hardware far from an Azure region. That's exactly where your most valuable assets, your AI models and your data, need protection the most. At Microsoft Ignite, we are excited to announce Confidential Computing on Azure Local.

Our commitment to confidential computing across the silicon and OEM ecosystem

We are committed to bring confidential computing to Azure Local across the latest platforms from AMD, Intel, NVIDIA, and our OEM ecosystem, giving customers greater choice in where and how they run protected workloads.

We are announcing the General Availability of Confidential Virtual Machines (CVMs) with 2610 Azure Local release on AMD platform. Intel TDX is already in advanced stages and preview expected Q1 of next year. The public preview of confidential GPUs on NVIDIA hardware, which extends hardware-based protection to accelerated AI workloads, is also coming soon.

AI models protection

Organizations invest significant resources in training AI models on their proprietary data. But when those models are deployed to remote sites, they're carrying valuable intellectual property into environments which may not be as secured as an Azure datacenter.

This is where Confidential VMs come in, following a simple principle: prove trust first, then unlock the intelligence. Your AI model is deployed in encrypted form and before the model can run, the CVM uses hardware-backed attestation to verify that it's running in a trusted environment. Only when that verification succeeds is the encryption key released, allowing the model to be unlocked and executed inside protected memory. If the environment fails verification, the key is never released, and the model remains encrypted and protected.

With GPUs now central to AI and other compute-intensive workloads, protecting sensitive data requires security that extends beyond the CPU. To address this, we're extending confidential computing protections to NVIDIA GPUs, enabling organizations to run demanding AI workloads while keeping model weights and sensitive data protected, even during GPU-accelerated processing.

Data protection

Azure Local also supports general-purpose confidential VMs, so organizations can run SQL databases, line-of-business applications and other data-intensive services inside a hardware-based trusted execution environment. Organizations can run workloads such as SQL databases inside Confidential VMs, keeping sensitive data protected even while it's being processed. Hardware-backed attestation also provides verifiable evidence of the trusted execution environment, helping organizations demonstrate security assurances to auditors and regulators.

Get started today

Ready to try it? Our Microsoft Learn documentation walks you through setting up a CVM-ready cluster and running your first confidential workload: https://learn.microsoft.com/azure/azure-local/manage/confidential-vm-overview

Interested in Intel TDX & AMD CVMs, or want to help shape what comes next? Share your input through our Interest and Readiness Survey https://aka.ms/Feedback-Confidential-Computing or email us at azlocal-cc@microsoft.com. We can't wait to see what you build.

 

 

Updated Nov 16, 2026
Version 1.0
No CommentsBe the first to comment