Nvidia wants to be the company that keeps AI agents on a leash. At its GTC Berlin keynote on October 21, CEO Jensen Huang unveiled the OpenShell Reference Architecture, a hardware-and-software blueprint for building, containing and monitoring AI agents — and a new kind of AI factory that can be installed in 14 days instead of six months.
OpenShell builds on Nvidia’s open-source OpenShell secure runtime. In the reference design, models run on Vera Rubin systems, agent harnesses and tool use run on Vera CPUs, and BlueField-4 chips run Nvidia’s Sentry software, which watches agent behavior from outside the main system and enforces policy in hardware. “As enterprises prepare to deploy billions of AI agents, safety and control must be built into the computing architecture,” Huang said. “Agents can think on Rubin, act on Vera and be governed on BlueField.”
The ecosystem is already lining up: Nvidia lists Anthropic, AWS, Google Cloud, Meta, Microsoft, OpenAI, Perplexity, SAP and SpaceX’s AI unit among those working with the architecture. The catch, per independent coverage: it requires Nvidia hardware, still has open security issues on GitHub, and collects telemetry by default unless users opt out.
The second reveal, DSX POD, targets the other bottleneck: compute capacity. These are complete Vera Rubin AI factories built by partners inside modular containers and validated before delivery, cutting onsite deployment from months to two weeks. Nvidia says DSX designs can eliminate up to 100% of cooling water use via closed-loop liquid cooling, and DSX MaxLPS software delivers the same AI output with up to 30% less power. CoreWeave, Google Cloud, Lambda, Nebius and Nscale plan to use the technology.
Builders shipping agentic software today can explore leading assistants in our directory, including GitHub Copilot, Cursor and Codeium.
Source: Tesla News Wire, Securities.io
