Briefs · Taipei time (GMT+8)

NVIDIA adds Confidential Bare Metal to its confidential computing lineup, letting model makers protect models and data on third-party servers

一句重點：Confidential Bare Metal now sits alongside Kubernetes containers and virtual machines as one of three ways to deploy. All three rest on the same security design, so a company can choose based on where it draws its own trust boundary (NVIDIA blog).

發生什麼事

The new option is about control. When a model maker rents bare-metal machines from a third-party cloud, a neocloud or a company's own data center, it can run even the host-level software itself. To make that possible, NVIDIA added several platform security features covering hardware-level access restrictions, measurement of platform integrity, and protection for data while it moves over the network (NVIDIA blog).

Hardware support is moving up as well. On the Rubin side it covers x86 HGX and DGX Rubin NVL8, while on the Blackwell side it reaches a full NVL72 rack. DSX OS, which manages AI factories, will later handle every confidential computing mode (NVIDIA blog).

01Sources

More briefs →