Keynote pitch for Europe

Jensen Huang addressed a crowd of 4,000 at the Tempodrom in Berlin on 21 October, during Nvidia's GTC Berlin conference. His vision combined robotics, AI agents and modular data centers, with Nvidia as the underlying technology. The March GTC in San Jose had focused on inference and efficiency, whereas this edition centred on power, agents and physical AI. Huang argued that Europe should produce AI rather than simply consume it.

“To be part of this AI economy, you don't want to just be consuming the AI. You want to be producing the AI.”

The DSX Pod and European partners

The headline product was the DSX Pod, a plug-and-play AI factory delivered pre-assembled in a large metal container with Nvidia's Vera Rubin compute racks. Nvidia says it can be online within 14 days, rather than the months or years a conventional build takes, which matters in Europe's slow-moving infrastructure market. Huang added that the pod would cut the time compute sits idle. Nvidia first presented DSX at GTC Taipei on 31 May 2026. On Wednesday it named European partners building on the platform, among them Nscale, Siemens, Schneider Electric, ABB, Siemens Energy, Hitachi Energy and Trane Technologies. Warren Barkley, who leads DSX product management at Nvidia, argued that European technology leadership rests on turning power into intelligence efficiently.

Efficiency claims and grid trials

Nvidia says the DSX tool MaxLPS can produce the same AI output with up to 30% less electricity. Nscale tested it on four racks of GB300 servers at its Iceland site, where inference throughput rose 49% within an unchanged 264-kilowatt budget. Closed-loop liquid cooling could cut water use to almost zero, Nvidia claims, and a two-gigawatt site could save 54 million litres a day. A UK trial with National Grid and Nebius found that a cluster of Nvidia chips cut its power draw by up to 40% during simulated grid events. Nvidia has not given dates for general availability. Nscale chief executive Josh Payne summed up the goal:

“Europe's AI ambitions demand that we make every megawatt count.”

MaxLPS electricity cut (up to)30%

Grid-event power draw cut (up to)40%

Inference throughput rise (Nscale Iceland test)49%

Sentry agent capacity gain (up to)20%

MaxLPS electricity cut (up to)

30 %

Grid-event power draw cut (up to)

40 %

Inference throughput rise (Nscale Iceland test)

49 %

Sentry agent capacity gain (up to)

20 %

Data from the article text

Agents, safety and robotics

Nvidia also released the OpenShell Reference Architecture, available now, for running AI agents inside fixed limits. It builds on OpenShell, an open-source runtime Huang said is released under the Apache 2.0 licence, and runs models on Vera Rubin systems. BlueField-4 units run Nvidia Sentry, which watches agents from outside and stops any attempt to leave their boundary within milliseconds, while raising agent capacity by up to 20%. Anthropic, OpenAI, Meta, Google and others can run agents on the infrastructure, and ServiceNow and SAP are building on it. Huang also called humanoid robotics the 'next great industry' and announced partnerships with VW and Mercedes on autonomous vehicles. He framed the security layer this way:

“AI capability and safety must advance together. As enterprises prepare to deploy billions of AI agents, safety and control must be built into the computing architecture.”

The protest and the Russian chip claims

Protesters interrupted the keynote about 20 minutes in, while Huang showed Vera Rubin hardware, per Nvidia's livestream. Two protesters stood on chairs holding a banner with the Greenpeace logo that told Nvidia to clean up its 'dirty business', and one asked what the company was doing about Nvidia chips found in Russian drones. Security led them out, Huang asked staff to go easy, and he returned to his slides. Greenpeace Germany and Greenpeace International had issued no statement by Wednesday afternoon. Ukraine says Jetson Orin modules have turned up in Russian weapons, and Tom's Hardware has reported Ukrainian claims that one sat in a Russian S-71 'Monochrome' cruise missile. Jetson Orin is a small computer module used in robotics and autonomous machines. Nvidia told Tom's Hardware its modules are consumer-grade products sold to students, developers and startups, and DCD reported that sanctions bar Russia from buying such hardware.