Overview
- NVIDIA announced the OpenShell Reference Architecture at GTC Berlin on October 21, 2026, and said the stack is available immediately for customers and partners.
- The architecture pairs Vera Rubin systems for model inference, Vera CPUs to run agent harnesses and BlueField‑4 DPUs running NVIDIA Sentry for continuous out‑of‑band monitoring and in‑silicon policy enforcement.
- NVIDIA says Sentry on BlueField‑4 can quarantine agents that try to leave their software boundary in milliseconds and claims up to a 20% boost in agent capacity, while Vera Rubin NVL72 and Vera CPUs are billed for major throughput and efficiency gains.
- A broad ecosystem of cloud providers, model builders, cybersecurity firms, enterprise software vendors and system makers — including ServiceNow, AWS, Google Cloud, Microsoft, Anthropic, CrowdStrike, Dell and HPE — are integrating OpenShell and tools such as ServiceNow’s AI Passport to carry and revoke runtime policies.
- Independent performance tests and widespread production deployments are still pending, so real‑world verification of NVIDIA’s throughput, CPU and enforcement claims will determine how widely regulated industries adopt the reference build.