Building Trusted AI Factories with NVIDIA Accelerated Confidential Computing
summary
NVIDIA presents an end-to-end stack that extends hardware root of trust into GPUs with isolation and encryption across CPU GPU and NVLink and adds composite attestation so model owners can move IP to hosted hardware. Reference architectures include confidential containers on Kubernetes plus confidential VMs, an OpenShell policy workspace for agent sandboxes, and an expanded NVIDIA Certified Systems program. Performance tests show near non‑CC throughput with framework optimizations and NVL72 trust boundary racks up to 72 GPUs and time to first token single‑digit overhead.
AI-written summary · about
chapters (12)
- 0:00NVIDIA and confidential computing
- 1:04AI shift to autonomous agents
- 3:12AI factories cost per token
- 5:23OpenShell for secure agents
- 7:51Confidential computing use cases
- 10:02Third-gen confidential GPUs
- 11:47NVIDIA Certified Systems
- 12:29Confidential containers with GPUs
- 14:35NVIDIA ecosystem partners
- 14:56Performance impact of confidential computing
- 18:55Vera Rubin confidential computing
- 19:58Closing and Q&A
related talks
20mTrusted Hardware for AI Workloads: Extending Confidential...
36mCTO Panel: The Impact of Confidential Computing with Intel, Azure, AMD & NVIDIA
47mTowards Confidential AI for the Masses!
42mSponsored Session: Confidential Computing - New Capabilities for New Workloads
44mGPU-less, Trust-less, Limit-less: Reimagining the Confidential AI Cloud
16mIntroduction to confidential computing
37mIt's All About Trust: Why Confidential Computing Matters- 42mSponsored Session: Confidential Computing - New Capabilities for New Workloads