Launching NVIDIA Confidential Bare Metal for Secure Deployment of Frontier Models
Public- 21 Oct, 2026

NVIDIA has announced the expansion of its Confidential Computing portfolio with the introduction of Confidential Bare Metal, a new deployment option designed for frontier model builders. This feature allows organizations to execute sensitive workloads on dedicated third-party infrastructure while retaining strict control over their proprietary models and data. As part of the broader NVIDIA OpenShell Reference Architecture, this initiative supports the safe deployment of agents at scale. Confidential Computing serves as the critical infrastructure layer that safeguards the models and data these agents depend on during execution, ensuring that sensitive operations remain protected within a secure environment across diverse hardware configurations.

The new platform introduces advanced security controls that protect data in transit, measure platform integrity, and enforce hardware-level access restrictions. These capabilities allow model builders to own and operate their bespoke software stacks, including host and platform infrastructure software, when using NVIDIA accelerated computing infrastructure. This flexibility is available across third-party cloud, neocloud, and enterprise infrastructure providers. By offering a converged security architecture with three distinct deployment options, NVIDIA enables organizations to match the appropriate security model to their specific trust boundaries and operational requirements. This ensures that high-value workloads receive the necessary privacy protections without compromising performance or operational autonomy.

Hardware support for this expanded ecosystem now includes NVIDIA Grace Blackwell NVL72 rack-scale systems, as well as x86-based NVIDIA HGX Rubin NVL8 and NVIDIA DGX Rubin NVL8 systems. The technology will also integrate with NVIDIA DSX OS, providing a reference architecture for cloud providers building AI factories. This integration ensures that cloud providers have a robust software foundation for operating confidential workloads. The expansion brings Confidential Computing to larger scale systems, allowing for more robust protection of training data and model intellectual property. This move significantly broadens the hardware landscape where secure, hardware-rooted isolation and cryptographic attestation can be applied to modern AI workloads.
Major industry partners are already integrating these new capabilities into their offerings. Cohere plans to integrate Confidential Bare Metal into its Model Vault, using trusted execution environments to extend hardware-rooted security directly to the GPU. This protects models and data in memory and across interconnects during execution. Google Cloud is extending support to A4X Max instances powered by NVIDIA GB300 NVL72 systems, combining non-virtualized accelerated computing with cryptographic protection. A wide range of providers, including CoreWeave, Microsoft Azure, and Vultr, support these deployments, while systems partners like Cisco, Dell, and HPE will offer NVIDIA-Certified Systems. This creates a comprehensive global ecosystem for scaling training and inference with strong privacy guarantees.