Nvidia wants to be the company that keeps AI agents on a leash. At its GTC Berlin keynote on Wednesday, Oct. 21, 2026, CEO Jensen Huang announced the OpenShell Reference Architecture for building, containing and monitoring AI agents, plus DSX POD, prebuilt AI factories that Nvidia says cut onsite deployment from six months to 14 days.

OpenShell: agents "contained, monitored and under control"

The architecture builds on Nvidia's open-source OpenShell secure runtime. Models run on Vera Rubin systems, agent harnesses and tool use run on Vera CPUs, and BlueField-4 chips run Nvidia's Sentry software to watch agent behavior from outside the main system and enforce policy in hardware, Nvidia said. It also extends to desk-side machines such as DGX Spark and DGX Station.

"As enterprises prepare to deploy billions of AI agents, safety and control must be built into the computing architecture," Huang said in the announcement. "Agents can think on Rubin, act on Vera and be governed on BlueField."

Nvidia lists Anthropic, AWS, Google Cloud, Meta, Microsoft, OpenAI, Perplexity, SAP and SpaceX Super Intelligence among those working with the architecture. The Register noted the catch: it requires Nvidia hardware, OpenShell still has open security issues on GitHub, and it collects telemetry by default unless users opt out. Banks BBVA and Santander are reviewing the spec, the Register reported.

DSX POD and greener AI factories

  • DSX POD: complete Vera Rubin AI factories built by partners inside modular containers and validated before delivery. "This is a complete transformation of the world supply chain," Huang said, per Nvidia's live blog.
  • Water and power: Nvidia says its DSX designs can cut cooling water use by up to 100% using closed-loop liquid cooling and dry coolers, and its DSX MaxLPS software can produce the same AI output with up to 30% less power. CoreWeave, Google Cloud, Lambda, Nebius and Nscale are among those planning to use it.

Bloomberg's live coverage reported that protesters briefly interrupted the keynote.

What it means for Tesla

Tesla and SpaceX are betting heavily on their own compute, from Colossus to the planned Terafab chip plant, and Musk has said power and speed are the hard parts of bringing compute online. Nvidia's 14-day claim is aimed squarely at that bottleneck. Our GTC Berlin preview has the replay details, and Nvidia's recent German physical-AI partnership set the stage.