Nvidia GTC Berlin: DSX POD, OpenShell Agent Controls
Nvidia unveiled DSX POD factory-built AI data centers and the OpenShell Reference Architecture at GTC Berlin, putting AI agent safety controls into hardware.
Nvidia CEO Jensen Huang used the GTC Berlin keynote on October 21, 2026 to announce two big platform moves: DSX POD, factory-built AI data centers that ship ready to run, and the OpenShell Reference Architecture, which bakes AI agent safety controls directly into hardware. Speaking at Berlin's Tempodrom, he argued that as companies prepare to deploy billions of AI agents, safety and control must be built into the computing architecture itself.
What happened
Huang's keynote at the GTC Berlin conference on October 21, 2026 centered on deployment speed and agent safety rather than new chips. The two headline announcements were DSX POD, a reference design for complete AI data centers built inside modular containers, and the OpenShell Reference Architecture, a layered system for running AI agents inside hardware-enforced limits. The event was briefly interrupted by environmental protesters holding banners, which Huang joked about before moving on. The keynote also covered robotics, cars, telecoms, and a 2.5x expansion of Nvidia's substrate supply capacity.
DSX POD: AI factories built in 14 days
DSX POD packs complete AI data centers into factory-built containers. Each unit combines seven liquid-cooled Vera Rubin compute racks with Nvidia's Spectrum-X networking, cabling, power, and liquid cooling. Partners cable, integrate, and test the systems inside modular containers before delivery, with worldwide shipments expected in early 2027. Nvidia says this cuts deployment at a prepared site from about six months to as little as 14 days. The 14-day figure assumes a prepared site and does not include the time needed to get power, connectivity, and site infrastructure ready first. Partners planning to deliver DSX POD systems in early 2027 include Dell Technologies, HPE, Lenovo, Foxconn, Schneider Electric, and Vertiv. Dell's version, the Dell Modular Data Center with Nvidia DSX POD, was announced the same morning. Huang also highlighted water savings: the systems use closed-loop liquid cooling with coolant temperatures up to 45 degrees Celsius, and he claimed that at a 2-gigawatt scale, DSX can save 54 million liters of water a day compared with conventional cooling.
OpenShell: agent safety enforced in silicon
The OpenShell Reference Architecture separates agent execution from hardware-level security enforcement. Models run on Vera Rubin systems, agent harnesses and tool use run on Vera CPUs, and BlueField-4 networking chips run Nvidia's Sentry software, which watches agent behavior from outside the main system and enforces policy in hardware. Nvidia says Sentry can quarantine and stop an agent within milliseconds if it moves outside its software boundary. "Agents can think on Rubin, act on Vera and be governed on BlueField," Huang said. "AI capability and safety must advance together." The architecture is available now and runs on Linux, Windows, and the major cloud platforms. Companies including Anthropic, OpenAI, Google, Meta, AWS, Microsoft, SAP, and SpaceX's AI unit are working with the design, alongside security vendors such as CrowdStrike and Palo Alto Networks.
Robots, cars, and phone calls
Huang spent much of the keynote on physical AI. On stage, small two-legged robot ducks from Pollen Robotics responded to his voice commands. He also unveiled the Agile ONE SN, a humanoid robot reference platform from German startup Agile Robots based on Nvidia's Isaac GR00T technology, scheduled for early 2027. In cars, Huang said Nvidia's Hyperion system has become an industry standard being designed into one fleet after another. In telecoms, he announced the first commercial AI-RAN phone call on T-Mobile's network with Nokia, saying "now CUDA makes phone calls." He also praised Europe's technology ambitions, noting Nvidia has invested almost $10 billion in Europe over the past year, including 14 investments in European startups during his current trip. A Forrester analyst cautioned that anything with critical ongoing dependencies on Nvidia is unlikely to meet the purest definitions of European tech sovereignty.
Why it matters
Nvidia is no longer selling just chips. DSX POD turns data center construction into a factory product, which could speed up AI infrastructure buildouts for companies and countries racing to add computing capacity. OpenShell addresses one of the biggest worries about AI agents: what happens when an autonomous system does something it should not. By enforcing limits in hardware rather than software, Nvidia is making a bet that safety will be a selling point for its full stack. The catch, as The Register noted, is that OpenShell requires Nvidia hardware, and many of the performance figures come from Nvidia without independent benchmarks. Still, with agent deployment accelerating, hardware-level guardrails are quickly becoming a competitive battleground.
Sources
- The Register - https://www.theregister.com/ai-and-ml/2026/10/21/jensen-huang-sends-in-the-robot-ducks-to-sell-nvidias-ai-everywhere-vision/5304472 - StorageReview - https://www.storagereview.com/news/nvidia-gtc-berlin-2026-dsx-pod-14-day-ai-factory-substrate-openshell-ai-ran?amp