aiqbench analysis
NVIDIA's new Confidential Bare Metal deployment allows model developers to execute sensitive workloads on third-party infrastructure while maintaining control over their models and data through hardware-level isolation. This expansion covers the latest Grace Blackwell, HGX Rubin, and DGX Rubin systems, with Google Cloud and Cohere adopting the solution for enhanced security in dedicated environments.
NVIDIA Launches Confidential Bare Metal to Expand AI Model and Data Confidential Computing Protection
NVIDIA has launched the Confidential Bare Metal deployment solution, enabling leading model developers to run sensitive workloads on third-party dedicated infrastructure while retaining control over their models and data. The solution ensures security during operation through hardware-level isolation, encrypted verification, and access restrictions. NVIDIA's confidential computing support now extends to Grace Blackwell NVL72, HGX Rubin NVL8, and DGX Rubin NVL8 systems, with integration into DSX OS. Google Cloud will offer this solution on A4X Max instances equipped with NVIDIA GB300 NVL72 systems, while Cohere plans to integrate it into its Cohere Model Vault.
Read the full article at 36KrRelated coverage

NVIDIA Confidential Computing Expands, Enabling Frontier Model Builders to Securely Scale Across Global Infrastructure
NVIDIA · Oct 21, 2026

Top 10 Best Cloud Encryption Solutions in 2026
Cybersecurity News · Sep 18, 2026

Enabling Private High-Performance Production AI Inference with NVIDIA Confidential Computing
NVIDIA · Sep 22, 2026

Principles for Embedded Evaluations
LessWrong · Sep 30, 2026