NVIDIA introduces confidential bare-metal infrastructure for AI

NVIDIA announced Confidential Bare Metal on October 21. The deployment option adds hardware-backed isolation and attestation for model builders using dedicated infrastructure operated by another provider.

The goal is to protect sensitive models and data during execution, alongside existing confidential-container and virtual-machine options. Availability depends on supported hardware and provider integrations; the announcement includes plans from cloud and infrastructure partners.

This addresses an infrastructure trust boundary. It does not determine whether a model's answers are accurate or whether an agent has been given appropriate access to a business system. Those controls remain separate parts of the application.