NVIDIA Launches Confidential Bare Metal for Secure Global AI Infrastructure

Summary

NVIDIA introduced Confidential Bare Metal on October 21, 2026, adding a third NVIDIA Confidential Computing deployment option alongside Confidential Containers for Kubernetes and Confidential Virtual Machines. It lets frontier model builders operate bespoke host and platform software on dedicated third-party cloud, neocloud and enterprise infrastructure while retaining control of models and data. New controls protect data in transit, measure platform integrity and enforce hardware-level access restrictions. The infrastructure complements the NVIDIA OpenShell Reference Architecture for safely deploying agents at scale.

Cohere plans to integrate Confidential Bare Metal into Cohere Model Vault, using trusted execution environments to protect models and data in GPU memory and across interconnects, CTO Phil Blunsom said. Google Cloud is extending support to A4X Max instances powered by NVIDIA GB300 NVL72 systems, combining non-virtualized accelerated computing with cryptographic protection for training data and model intellectual property, distinguished engineer Andrés Lagar-Cavilla said. Hardware-rooted isolation and cryptographic attestation now extend to Grace Blackwell NVL72, x86-based HGX Rubin NVL8 and DGX Rubin NVL8 systems. CoreWeave, Firmus, GMI Cloud, Google Cloud, IREN, Lambda, Microsoft Azure Infrastructure, Nebius, Nscale and Vultr support deployments. Cisco, Dell Technologies, HPE, Lenovo, Supermicro and Bull will support them through NVIDIA-Certified Systems. NVIDIA DSX OS will support all three modes as a reference architecture for confidential AI factories, expanding geographic, sovereign and operational choices for global training and inference.