Skip to content
Brief Security & Trust ·

NVIDIA adds Confidential Bare Metal for frontier AI model builders

NVIDIA said frontier model builders can now run private workloads on dedicated servers owned by others.

In brief
  • NVIDIA added Confidential Bare Metal for frontier model builders. They can run sensitive work on dedicated servers from other companies.
  • Builders can run their own host and platform software. NVIDIA gave no prices, start dates or test results.

NVIDIA said on 21 October 2026 that it is adding Confidential Bare Metal to its Confidential Computing line. The option is for frontier model builders. They can run sensitive work on dedicated servers from other companies and still control their models and data. Google Cloud said it will support it on A4X Max instances built on GB300 NVL72 systems. Cohere said it plans to use it in Cohere Model Vault.

NVIDIA says the system relies on hardware-based isolation and cryptographic attestation, a signed proof of what is running, to protect models and data during use. Cohere said this reaches memory and the links between chips. Bare metal means a whole physical server with no virtual machine layer. NVIDIA said new controls protect data in transit, check platform integrity and limit access in hardware. NVIDIA lists ten providers as backers of its confidential deployments, including CoreWeave, Nebius and Vultr. It gave no prices, start dates or test results, and did not say which providers will offer bare metal.

The announcement moves the trust line. NVIDIA says builders can run their own host and platform software, so the provider supplies the servers while the builder controls the stack. Buyers should ask which providers will offer it, how attestation is verified, and what it costs and when.

A WebPulse Brief: a short report of an important event, written by the WebPulse Newsroom with AI assistance and checked against the reporting below. When there is more to explain, we follow up with a full story. How we use AI.

Reporting: NVIDIA.