NVIDIA introduced Confidential Bare Metal, extending its confidential-computing options for model builders using cloud, neocloud and enterprise infrastructure.
NVIDIA announced Confidential Bare Metal on October 21, 2026, adding a deployment option intended to protect sensitive AI workloads running on dedicated third-party infrastructure. The company says model builders can operate their own software stack—including host and platform software—while retaining control of models and data on NVIDIA accelerated-computing systems.
The new option adds platform security controls for data in transit, platform-integrity measurement and hardware-level access restrictions. NVIDIA describes the aim as protecting models and data during execution, including in memory and across interconnects. Confidential Bare Metal joins two existing deployment approaches in NVIDIA Confidential Computing: Confidential Containers for Kubernetes and Confidential Virtual Machines for virtualized environments. Organizations can select among the three according to their operating needs and trust boundaries.
NVIDIA said the capabilities are expanding across its platforms, including Grace Blackwell NVL72 rack-scale systems and x86-based HGX Rubin NVL8 and DGX Rubin NVL8 systems. The company also plans integration with NVIDIA DSX OS, which it positions as a reference architecture for cloud providers building AI factories.
Several providers and model builders were named. Cohere plans to integrate Confidential Bare Metal into Cohere Model Vault, with the stated goal of extending hardware-rooted protections to GPU execution. Google Cloud is extending support to A4X Max instances powered by NVIDIA GB300 NVL72 systems. NVIDIA listed CoreWeave, Firmus, GMI Cloud, IREN, Lambda, Microsoft Azure Infrastructure, Nebius, Nscale and Vultr among providers supporting Confidential Computing deployments; it also named Cisco, Dell Technologies, HPE, Lenovo, Supermicro and Bull as systems partners supporting the capabilities on NVIDIA-Certified Systems.
The announcement places confidential-computing controls alongside NVIDIA’s broader stack for deploying AI agents, including the OpenShell Reference Architecture mentioned in the article. For organizations considering external compute capacity for proprietary models or data, the new bare-metal configuration is an additional infrastructure choice rather than a replacement for containerized or virtualized options.