At GTC Berlin, Nvidia launches DSX POD AI factories and an OpenShell reference architecture to keep AI agents inside set boundaries

Jensen Huang gave Nvidia's GTC Berlin keynote on 21 October. Nvidia's blog of the keynote says its Vera Rubin systems are “in full production”. It describes DSX POD as a reference design for complete Vera Rubin AI factories that partners build in modular units, and says: “With DSX POD, onsite deployment can shrink from six months to just 14 days once site utilities are ready.” Nvidia also says it is “committing nearly $10 billion to the technology ecosystem across Europe”.

For agents, Nvidia announced the OpenShell Reference Architecture, built on its OpenShell secure runtime. It has three parts: Vera Rubin NVL72 systems for model inference, Nvidia Vera CPUs running OpenShell for agent harnesses and tool use, and BlueField-4 DPUs running Nvidia Sentry to enforce the rules in hardware. Nvidia lists about 19 cloud partners supporting it, among them CoreWeave, Crusoe, Lambda, Nebius and Nscale. It also expanded its open-source Agent Toolkit with an Agent Knowledge Foundation, an Auto-Ontology library that maps company data to business concepts, and Kumo Structured open models (Tabular, Relational, Time Series and Forecast). Nvidia writes: “OpenShell provides the secure runtime boundary, giving enterprises control over what agents can access and do.”

These are Nvidia's own claims, and it gives no prices. One headline the same day says Google Cloud and Nvidia announce general availability of Vera Rubin instances; the post itself says general availability “is coming next month”. The Switchboard covered Nvidia's agent watchdog on the network card on 28 September; what is new is the full reference architecture and its partner list.

• Claimed Nvidia says DSX POD can shrink on-site deployment of an AI factory from six months to 14 days once site utilities are ready. NVIDIA Blog, GTC Berlin keynote
• Claimed The OpenShell Reference Architecture pairs Vera Rubin NVL72, Vera CPUs running OpenShell and BlueField-4 DPUs running Nvidia Sentry; Nvidia names about 19 supporting cloud partners. NVIDIA Blog, cloud partners in Europe
• Claimed Nvidia's Agent Toolkit adds an Agent Knowledge Foundation, an Auto-Ontology library and Kumo Structured open models. NVIDIA Blog, Agent Toolkit
• Confirmed Google Cloud's Vera Rubin A5X instances are not yet generally available: the post says general availability is coming next month. NVIDIA Blog, Google Cloud and Vera Rubin