Google Cloud expanded its Confidential VM offerings with several general availability milestones in 2026, strengthening hardware-based memory encryption for data in use. The updates, detailed in release notes published through September, cover Intel TDX on c4-standard and c3-standard-lssd machine types plus accelerator-optimized instances for AI and machine learning tasks.
These changes build on existing AMD SEV and SEV-SNP support. Organizations handling regulated data or sensitive AI training now gain additional options for keeping memory contents encrypted even from the hypervisor.
Intel TDX Support and Accelerator Configurations Reach General Availability
Intel TDX became generally available on c4-standard-* machine types on September 23, 2026, and on c3-standard-*-lssd instances on September 2. The feature uses Intel Trust Domain Extensions to isolate VM memory with keys stored in dedicated hardware. Attestation continues through Google’s virtual Trusted Platform Module.
On August 7, Google Cloud made the g4-standard-48 machine type generally available. It combines fifth-generation AMD EPYC Turin processors with AMD SEV and one NVIDIA RTX PRO 6000 GPU. This configuration supports confidential AI workloads where both CPU and GPU memory remain protected. Google Cloud also extended Confidential GKE nodes with NVIDIA Confidential Computing integration, allowing customers to run containerized AI jobs under the same hardware protections.
Additional capacity improvements arrived earlier in the year. AMD SEV on C3D and C4D instances now supports more than 255 vCPUs following July updates. Live migration reached general availability on qualifying C3D AMD SEV instances in the first half of 2026, reducing operational friction for long-running confidential workloads.
Security Maintenance and Integration with Existing Google Cloud Services
Google Cloud addressed Intel TDX firmware issues through fleet-wide updates issued under GCP-2026-053 in August 2026. Customers typically require no action. AMD SEV-SNP instances experienced temporary boot and performance effects during kernel updates between August and November 2026.
Confidential Space received new container images on October 5, 2026, and gained general availability support for H100 GPUs with Intel Trust Authority attestation in September. These images integrate with Vertex AI, Dataflow, and GKE, extending encryption-in-use controls alongside the at-rest and in-transit protections already standard in Compute Engine.
The updates target financial services, healthcare, and other sectors that must demonstrate control over data while it is processed. Customers can select On-Demand, Spot, or Reservations pricing for the new confidential configurations, matching the flexibility available for non-confidential VMs.
Practical Steps for Adopting the Updated Confidential VMs
Teams evaluating the changes should first verify that their workloads fit the supported machine types and regions. Documentation lists exact vCPU and memory limits for each confidential configuration. Existing non-confidential workloads can be migrated by changing the machine type and enabling the confidential option during instance creation or through live migration where supported.
Organizations already using Vertex AI or GKE can test confidential nodes by selecting the appropriate node pool settings. Performance testing remains advisable, especially for AI training jobs that previously ran on standard GPU instances. Google Cloud continues to publish updated release notes and security bulletins as further refinements appear.
Frequently asked questions
Which machine types now support Intel TDX in Confidential VMs?
Intel TDX is generally available on c4-standard-* instances since September 23, 2026, and on c3-standard-*-lssd instances since September 2. These options add to existing AMD SEV support.
How do Confidential G4 VMs protect GPU workloads?
The g4-standard-48 instances combine AMD SEV with an NVIDIA RTX PRO 6000 GPU and NVIDIA Confidential Computing integration. Both CPU and GPU memory stay encrypted during processing.
Do the updates require customer action for security fixes?
Fleet-wide Intel TDX firmware updates were applied by Google Cloud in August 2026. Customers do not need to take action in most cases, though AMD SEV-SNP instances may show temporary performance effects during kernel updates.
Which Google Cloud services integrate with the new confidential configurations?
Confidential VMs and nodes work with GKE, Vertex AI, and Dataflow. Customers can apply the same encryption controls across compute, containers, and data processing pipelines.
By YasTech Editorial
Comments 3