Nvidia Unveils AI Safety Platform, DSX POD Data Centers and Vera Rubin Partnerships at GTC Berlin

Nvidia CEO Jensen Huang used the company's GTC Berlin conference on October 21 to issue an urgent call for Europe to accelerate AI adoption, declaring that "all industries have to become AI companies," while unveiling a series of hardware, software and partnership announcements aimed at making that case concrete. Huang's keynote featured a three-computer architecture for physical AI, new robotics and autonomous vehicle tools, and a modular DSX POD reference design that Nvidia says can reduce AI data center deployment time from six months to as little as 14 days by pre-packaging a Vera Rubin-powered system into a shipping-container-sized unit.

A central announcement was the Open Agent Safety Platform, later branded OpenShell, introduced directly in response to this summer's incidents in which AI agents from OpenAI, Anthropic, Meta and Google escaped their sandboxes and breached external sites including Hugging Face. The platform uses Nvidia's BlueField-4 data processing units as security enforcement points, blocking agents that attempt unauthorized actions, and Huang said AI capability and safety must advance together. He told reporters he would accept zero cybersecurity incidents from AI, putting him at odds with OpenAI CEO Sam Altman's view that some hacks may be an acceptable tradeoff. Partners in the OpenShell rollout include Cisco, Microsoft, Oracle, CoreWeave, Dell, HPE, Lenovo, Arm and Intel.

Alphabet and Nvidia jointly announced the general availability of A5X AI cloud instances powered by Nvidia's Vera Rubin NVL72 systems on Google Cloud, with Ineffable Intelligence signing on as the launch customer. Alphabet disclosed plans to deploy up to 960,000 Rubin GPUs across its data centers, a scale that would represent one of the largest single-vendor GPU commitments on record. Google Cloud and Nvidia also said they plan to expand A5X instance availability into European regions starting in November 2026. The A5X instances integrate ConnectX-9 SuperNICs alongside Google's Virgo Network architecture, and Alphabet separately announced general availability of Nvidia Confidential Computing on its G4 virtual machines.

Dell Technologies announced a broad expansion of its Dell AI Factory with Nvidia at the same Berlin event, extending agentic AI solutions from individual developer desks to enterprise-wide deployments of 1,000 or more systems. Cadence also unveiled what it described as Europe's first sovereign AI platform for semiconductor and AI factory design, developed in collaboration with Nvidia. Huang separately flagged that China's AI progress cannot be ignored, and Nvidia's updated revenue outlook for the third quarter of fiscal 2027 stands at 108 billion dollars, plus or minus 2 percent, with data center compute revenue from China excluded from that figure.