NVIDIA Confidential Bare Metal Expands Secure AI Infrastructure for Frontier Models

NVIDIA is expanding its confidential computing portfolio with Confidential Bare Metal, a new deployment option that helps frontier model builders run sensitive AI workloads on dedicated third-party infrastructure while maintaining control over their models and data.

NVIDIA is building tools to enable secure AI agents across the stack. The NVIDIA OpenShell Reference Architecture, announced today, is part of NVIDIA’s growing technology portfolio for securely deploying agents at scale. NVIDIA Confidential Computing provides the infrastructure layer that helps protect the models and data those agents depend on while they are running.

Confidential Bare Metal extends that layer with new platform security controls designed to protect data in transit, measure platform integrity and enforce hardware-level access restrictions. These controls enable frontier model builders to own and operate bespoke software stacks—including host and platform infrastructure software—on bare metal instances running on NVIDIA accelerated computing infrastructure across third-party clouds, neoclouds and enterprise environments.

Three NVIDIA Confidential Computing deployment options

NVIDIA Confidential Computing now offers a unified security architecture with three deployment options:

• Confidential Bare Metal: For bare metal infrastructure
• Confidential Containers: For Kubernetes environments
• Confidential Virtual Machines: For virtualized environments

Organizations can choose the model that best fits their trust boundaries and operational requirements.

NVIDIA Confidential Computing also extends to NVIDIA Grace Blackwell and Rubin systems. It integrates with NVIDIA DSX OS as a reference architecture for cloud providers building AI factories.

Confidential Bare Metal gives model builders more deployment choices

Confidential Bare Metal provides supported providers with a way to deliver high-value workloads that require strong model and data privacy protections. For model builders, the expanding ecosystem offers more geographic, sovereign and operational options for scaling AI training and inference around the world.

Cohere Model Vault uses NVIDIA Confidential Computing

Cohere plans to integrate Confidential Bare Metal. Cohere Model Vault uses the trusted execution environment provided by NVIDIA Confidential Computing to extend hardware-based security directly to the GPU, helping protect models and data in memory and across the interconnect during execution.

“Enterprises trust Cohere to provide a secure and confidential inference platform,” said Phil Blunsom, chief technology officer at Cohere. “With NVIDIA Confidential Computing in Cohere Model Vault, we extend hardware-based security directly to the GPU, protecting customer data in memory and across the running interconnect.”

Google Cloud expands Confidential Bare Metal support

Google Cloud is extending Confidential Bare Metal support to A4X Max instances with NVIDIA GB300 NVL72 systems. These instances combine non-virtualized NVIDIA accelerated computing with cryptographic protection for training data and model intellectual property.

“Google Cloud introduced Confidential Bare Metal to provide AI labs with a hardware-isolated, non-virtualized computing environment and end-to-end cryptographic protection. Our A4X instances built using NVIDIA Confidential Computing maintain the privacy of user data, protect model IP and provide the highest level of service performance,” said Andrés Lagar-Cavilla, a distinguished engineer and AI infrastructure security lead at Google Cloud.

NVIDIA Confidential Computing expands across GPU platforms

NVIDIA Confidential Computing uses hardware-based isolation and cryptographic attestation to help protect models and data while they are running.

Support has been extended to x86-based NVIDIA HGX Rubin NVL8 and NVIDIA DGX Rubin NVL8 systems, in addition to the NVIDIA Grace Blackwell NVL72. This brings confidential computing to NVL72 rack-scale systems.

Cloud and system partners support confidential AI deployments

Cloud providers supporting the deployment of NVIDIA Confidential Computing include:

• CoreWeave
• SOLID
• GMI Cloud
• Google Cloud
• Airen
• Lambda
• Microsoft Azure infrastructure
• Nevius
• N Scale
• Bartol

Global system partners supporting these capabilities include Cisco, Dell Technologies, HPE, Lenovo, Supermicro and Bull. These features are supported by NVIDIA Certified Systems.

NVIDIA DSX OS supports all NVIDIA Confidential Computing modes and provides cloud providers with the software foundation to build and operate AI factories that support confidential workloads.

Learn more about NVIDIA Confidential Computing.