Protecting the workloads that matter most
Useful AI depends on the records and expertise that organizations work hardest to protect such as patient information, financial histories, proprietary models and the context an agent uses to act. Customers want to put that information to work without exposing it more widely. They also need to bring existing sensitive applications forward, choose infrastructure that fits the workload, and keep those systems reliable as they grow.
At Ignite 2026, Azure Confidential Computing is advancing both needs. New CPU and GPU options give customers more ways to protect records and models during processing. Confidential container capabilities make that protection easier to adopt in application platforms, while migration, encryption and recovery improvements help teams move and operate sensitive workloads with less disruption.
Here are the highlights:
- A confidential bare-metal technology showcase on NVIDIA Vera Rubin NVL72, with a single-node, four-GPU AI demonstration.
- A gated preview of 5th generation AMD EPYC™ (“Turin”) processor v7 confidential VMs, with more general-purpose and memory-optimized choices for sensitive workloads.
- Intel TDX v6 platform improvements and a protected-I/O technology showcase for next-generation Confidential VMs with TDX Connect and Azure Boost Confidential Device.
- Generally available migration and large OS-disk encryption capabilities, with further progress in backup and confidential data-disk protection.
- Intel TDX support for confidential AKS worker nodes (generally available with Azure Container Linux and in preview with Ubuntu 26.04), alongside engineering investments in next-generation Confidential Child Capable VMs, confidential enclaves, and pod-level confidential containers on AKS, plus progress in confidential databases and controlled collaboration.
- A Project Nova-based demonstration using zero-knowledge proofs to verify that a financial transaction meets agreed conditions without revealing the underlying balances or sensitive records.
Confidential AI: verify before sharing
Confidential AI needs to protect both sides of an interaction: the customer’s prompts and private context, and the model owner’s weights and intellectual property. Hardware-backed trusted execution environments protect data in use. Attestation provides signed evidence about the environment; authorization and key-release policies decide whether sensitive assets should become available to it.
Confidential computing addresses protection during processing. Neither replaces application permissions, careful handling of logs and tool calls, or the governance needed for a particular deployment.
Our work with NVIDIA on Vera Rubin NVL72, or VR72, starts with confidential bare-metal instances. The Ignite technology showcase uses one compute node with four Rubin GPUs to run a 397-billion-parameter model through vLLM. It collects CPU and GPU evidence and uses a minimal local verifier to check signatures, token binding and expected GPU measurements before demonstrating inference.
In this bare-metal model, the tenant’s host operating system and runtime are inside the trusted computing base. Customers control that stack and must secure it accordingly.
Figure 1. The confidential bare-metal architecture includes the tenant host OS in the trusted computing base; UEFI and BMC are outside it. Remote attestation is an architectural direction, while the Ignite demonstration uses a local verifier. Source: “Confidential AI at rack scale,” C-BMI architecture figure.
VR72 confidential VMs, AKS and Kata-based confidential containers, and Microsoft Foundry integration remain in development. Customers building today can start with Azure’s existing NVIDIA H100 confidential GPU VMs while evaluating the isolation, attestation and operating model their workload needs. Our reference architecture for protecting sensitive AI with NVIDIA H100 and Protopia offers another way to explore that design.
More choice for confidential compute
AMD EPYC™ (“Turin”) processor v7 CVM: more confidential VM choices
The gated preview of our next-generation confidential VMs brings 5th generation AMD EPYC™ (“Turin”) processor to six series: DCasv7, DCadsv7, DCalsv7, DCaldsv7, ECasv7 and ECadsv7. General-purpose and memory-optimized options, different memory-to-vCPU ratios, and configurations with or without local disks give customers more room to match infrastructure to the workload.
The platform combines AMD SEV-SNP with Azure Boost, Microsoft Azure Network Adapter (MANA), and NVMe support for local and remote storage. Secure AVIC adds protection for interrupt handling in SEV-SNP guests. Together, these changes target the compute, memory and I/O demands of sensitive enterprise applications without making one benchmark stand in for every workload.
For details, please refer to https://aka.ms/cvm-turin-preview.
Intel TDX: v6 improvements and protected-I/O showcase
For Intel TDX v6 confidential VMs, improvements include NVMe remote storage, posted interrupts, timer virtualization and local sealing. Customers should compare performance against a matched configuration and their own application behavior. Storage, networking and workload concurrency all matter; there is no single confidentiality-overhead number that describes every deployment.
For the next generation of confidential VMs, we are demonstrating protected I/O with Intel TDX Connect and Azure Boost Confidential Device (ABCD). Device attestation, TEE Device Interface Security Protocol (TDISP) and encrypted links extend the design beyond CPU memory to the devices a confidential workload uses. This technology demonstration is separate from the current v6 improvements; it is not a general-availability or public-enrollment announcement. We also continue to make progress on live migration for Intel TDX, building on the demonstration at Build.
Azure Local confidential computing
Confidential VMs powered by AMD SEV-SNP are now generally available on supported AMD EPYC Genoa processors, providing hardware-backed memory encryption and attestation to help protect sensitive workloads while running. We're also expanding hardware choice with Intel TDX-based confidential VMs in preview on Intel Xeon processors.
We're soon extending confidential computing to GPU-accelerated AI workloads with preview support for confidential GPUs on Azure Local. This enables organizations to run sensitive AI inference workloads with additional hardware-backed protection for data in use, helping bring confidential AI capabilities to customer-controlled infrastructure.
From confidential AKS nodes to protected application workloads
Platform teams running Kubernetes on Azure already use confidential AKS worker nodes on AMD SEV-SNP to run a worker node inside a hardware-backed VM boundary. At Ignite, we are expanding confidential AKS worker nodes to Intel TDX, now generally available with Azure Container Linux and in preview with Ubuntu 26.04. In the worker-node model, pods scheduled onto a node share that VM's protection boundary, which fits clusters where workloads and node administrators belong to the same trust domain.
When a specific service handles signing keys, credentials, or regulated records that must stay isolated from the surrounding host OS and node operators, the trust boundary needs to shrink to the workload itself. Today, confidential containers on Azure Container Instances provide a managed container option for workload-scoped, attested execution and policy enforcement. We are also building the next generation of Confidential Child Capable VMs on AMD EPYC processors with AMD SEV-SNP. In this model, a customer-operated parent VM manages lifecycle, scheduling, I/O, and availability, while a hardware-isolated child environment helps protect sensitive code and memory and helps keep the parent OS and parent-level administrators outside the child's memory confidentiality boundary within AMD's documented threat model.
We are building two workload experiences on top of these upcoming Confidential Child Capable VMs: confidential enclaves, a feature for splitting a single sensitive component out of a parent VM, and next-generation support in confidential containers on AKS (built on Kata Containers and the Confidential Containers project) for per-pod isolation. Both capabilities are under active development and planned for a future release.
Make confidential workloads easier to operate
Protection during execution is only useful if organizations can also migrate, recover and maintain the workload. Two generally available capabilities remove important migration barriers: Azure Migrate support for eligible moves to confidential VMs, and Large OS Disk Encryption for confidential VM migrations.
Large OS Disk Encryption supports OS disks larger than 127 GB, including customer-managed-key configurations. Supported combinations cover Windows, Ubuntu, Red Hat Enterprise Linux and Rocky Linux. Teams still need to check their source platform, guest image and target configuration. Source machines and replication paths need their own controls; confidential protection begins when the target runs in the supported trusted execution environment and meets its key-release conditions.
Azure Backup for confidential VMs is in public preview for supported v6-series VMs with OS-disk encryption, using platform-managed or customer-managed keys. Use the current Azure Backup support matrix to plan restore behavior rather than assuming parity with every standard VM backup feature.
Confidential Data Disk Encryption, in private preview, addresses another part of the lifecycle. It encrypts attached data disks inside the confidential guest before data reaches the host, with customer-managed keys released under authorization and attestation policy.
A protected data foundation for AI and collaboration
Models are only one part of an AI system. Databases hold the business records, retrieved context and agent memory that make those systems useful. SQL Server on Azure confidential VMs and Azure Local gives customers control of the database and guest software stack, together with responsibility for patching, backup and availability. Azure Database for PostgreSQL flexible server provides a managed confidential-computing option within its documented processor and regional support.
We are working to broaden database deployment choices. For now, customers should plan against the current PostgreSQL support matrix rather than assume every confidential VM family or region is available through the managed service. A managed database has a wider, Microsoft-managed trust boundary. PostgreSQL’s documented offering does not expose remote attestation to customers, so its control model should not be equated with a customer-managed confidential offering.
Azure Confidential Clean Rooms, in limited preview, applies this thinking across organizations. Participants can securely analyze privacy sensitive datasets from multiple collaborators with confidential compute-enabled Spark SQL-based big data analytics protecting their raw data from other collaborators and from the Azure operator. Campaign reach and measurement, and collaboration in financial services, illustrate the opportunity.
Prove the fact without sharing the record
Confidential computing protects sensitive information while it is processed. Zero-knowledge proofs provide a complementary capability, enabling independent verification of specific claims without revealing the underlying data. Across regulated industries, organizations need to establish trust while protecting confidential information. Project Nova explores how zero-knowledge proofs can enable privacy-preserving transactions, interoperability, and regulatory assurance across organizational boundaries.
In financial services, zero-knowledge proofs could enable cross-chain settlement by verifying transaction conditions across networks, support selective disclosure of sensitive financial information, and provide regulators with independent assurance of compliance without exposing underlying records. Beyond financial services, healthcare organizations could verify patient eligibility without sharing medical histories; AI systems could prove defined computations or policy checks without exposing sensitive inputs; and supply chains could demonstrate compliance or product provenance without revealing confidential supplier relationships. Project Nova aims to enable trusted verification with greater privacy and less unnecessary data sharing.
ACC portfolio of services and capabilities
ACC spans infrastructure, containers, data, AI, and foundational security services, giving customers multiple ways to protect sensitive workloads while they are in use. The portfolio is designed to meet customers where their workloads run today, while extending confidential computing to emerging AI and sovereignty scenarios.
Get started and go deeper
Start with the Azure Confidential Computing portfolio in our docs pages to find the deployment model and supported configuration that fit your application. Test attestation, key release and recovery together, follow the Azure Confidential Computing blog for deeper product guidance and see our GitHub repo for code and sample applications that take advantage of our technology.