NVIDIA expands confidential computing with bare-metal option
The new bare-metal option gives frontier model builders dedicated third-party infrastructure with hardware-rooted isolation. Cohere and Google Cloud detailed plans to adopt it.
NVIDIA has introduced Confidential Bare Metal, a new addition to its Confidential Computing portfolio. The company said the option lets frontier model builders run sensitive work on third-party bare-metal servers while keeping control of their models and data. [1]
According to NVIDIA, the move creates one security architecture with three deployment choices: Confidential Bare Metal for bare-metal systems, Confidential Containers for Kubernetes, and Confidential Virtual Machines for virtualized environments. Customers can pick the model that matches their trust boundaries and operating requirements. The company said confidential computing protects models and data during execution through hardware-rooted isolation and cryptographic attestation. [1]
The new controls cover data in transit, platform integrity checks, and hardware-level access enforcement. With them, frontier model builders can run their own software stack, including host and platform components, on NVIDIA-accelerated infrastructure in third-party clouds, neoclouds, and enterprise data centers. [1]
NVIDIA said the new option lets supported providers serve high-value work that demands strong model and data privacy. The company said model builders get more options across geography, sovereignty, and operations for scaling training and inference globally. [1]
Cohere plans to add Confidential Bare Metal to its Model Vault product. The company said it will use NVIDIA trusted execution environments to bring hardware-rooted security to the GPU, protecting data in memory and across interconnects while models run. Phil Blunsom, Cohere's chief technology officer, said, "we're extending hardware-rooted security directly to the GPU." [1]
Google Cloud said it will offer Confidential Bare Metal on A4X Max instances that run on NVIDIA GB300 NVL72 systems. These instances will pair non-virtualized NVIDIA computing with cryptographic protection for training data and model intellectual property. Andrés Lagar-Cavilla, distinguished engineer and AI infrastructure security lead at Google Cloud, said, "Google Cloud introduces Confidential Bare Metal to provide AI labs with hardware-isolated, non-virtualized compute environments and end-to-end cryptographic protection." [1]
NVIDIA said support now covers Grace Blackwell NVL72 rack-scale systems and x86-based HGX Rubin NVL8 and DGX Rubin NVL8 systems. The company listed Google Cloud, Microsoft Azure Infrastructure, CoreWeave, Lambda, Nebius, Nscale, IREN, Firmus, GMI Cloud, and Vultr as providers supporting deployments. Systems builders Cisco, Dell Technologies, HPE, Lenovo, Supermicro, and Bull will support the capabilities on NVIDIA-Certified Systems. [1]
NVIDIA said DSX OS will support every Confidential Computing mode, giving cloud providers a software base for AI factories that handle confidential workloads. [1]
Sources
Noor Haddad covers attacks, breach disclosures and exploited vulnerabilities for Five Nines. Noor writes from CISA and vendor advisories, SEC disclosures and published reporting.
Corrections. Report an error to corrections@thefivenines.com. We list substantive corrections on the corrections page.