ABAB News Logo
Back to Flash News

Jensen Huang’s Berlin GTC summary: Europe is building an ecosystem from AI factories to robots

On October 26, 2026, NVIDIA CEO Jensen Huang posted on X reviewing GTC Berlin, calling himself “the opening act for a family of ducks” and thanking Berlin. He said Europe is building an AI ecosystem and economy spanning AI factories, agents, and robots. He specifically mentioned DSX PODs (a Vera Rubin modular factory delivered after being built, wired, and tested inside the factory, and able to directly produce intelligence), Agent Toolkit (integrating models, harnesses, sandboxes, knowledge, and tools so enterprises can control what agents know, can access, and can do, with OpenShell enhancing security), new Physical AI customers (covering AI-RAN, robotaxis, and robots), and RTX Spark bringing personal agents to Windows PCs after four years of cooperation with Microsoft.

He thanked French President Emmanuel Macron, British Prime Minister Andy Burnham, the German chancellor, and others for hosting him, and mentioned celebrating Sferical One with Swedish technology leaders, the welcome from Kazakh President Tokayev, as well as partners and the NVIDIA team. A link to the full keynote was attached. The keynote was held in Berlin on October 21, and NVIDIA’s official blog confirmed DSX POD as a complete Vera Rubin AI factory reference design, delivered after partners integrate and validate it inside modular containers; once deployed on site after the facility is ready, deployment time can be shortened from about six months to 14 days.

The duck-related reference points to Pollen Robotics’ Microducks appearing on the keynote stage. In the keynote, Huang described the AI economy as a five-layer structure (energy, computing, AI factories, models, applications) and emphasized Europe’s path from consumers to producers. The Vera Rubin platform has entered full production, and RTX Spark is a recently launched personal AI device.

Original sources:

  1. Title: I was the opening act for a family of ducks... Publisher: Jensen Huang (@JensenHuang) Publication date: 2026-10-26 URL: https://x.com/JensenHuang/status/2114773600574279837 Source nature: post by the person involved Verification status: The full post was actually read in this report, verification date 2026-10-26 Supported content: GTC Berlin review, DSX POD description, Agent Toolkit and OpenShell, Physical AI customers, RTX Spark, statement about Europe building, and thanks to leaders.

  2. Title: GTC Berlin 2026: Live Updates From NVIDIA CEO Jensen Huang’s Keynote Publisher: NVIDIA Blog Publication date: 2026-10-21 URL: https://blogs.nvidia.com/blog/gtc-berlin-2026-keynote/ Source nature: official announcement Verification status: Relevant passages were actually read in this report, verification date 2026-10-26 Supported content: DSX POD as a modular Vera Rubin AI factory reference design, delivery after in-factory integration and validation, onsite deployment shortened from six months to 14 days, Agent Toolkit examples, RTX Spark personal AI.

  3. Title: Jensen Huang's GTC Berlin Keynote: NVIDIA Unveils DSX POD, OpenShell, and Physical AI Platforms Publisher: TAO Media Publication date: 2026-10-21 URL: https://www.tao.media/jensen-huangs-gtc-berlin-keynote-nvidia-unveils-dsx-pod-openshell-and-physical-ai-platforms/ Source nature: media report Verification status: This report was actually read, verification date 2026-10-26 Supported content: announcement of DSX POD and OpenShell, five-layer AI economy model, Physical AI platform, RTX Spark demonstration.

ABAB AI Insight

Jensen Huang’s Berlin summary positions NVIDIA as moving from a chip supplier to a provider of complete AI factory and agent infrastructure. DSX POD shifts integration, wiring, and validation into factory containers, aiming to cut onsite deployment time from months to about two weeks, provided the facility is ready, reducing friction for building AI factories in Europe and elsewhere. Vera Rubin has entered full production, supporting scaling from models to full-agent processing. This continues the company’s path of “extreme co-design,” bundling hardware and software delivery.

At the agent layer, Agent Toolkit integrates models, harnesses, sandboxes, knowledge, and tools, while OpenShell provides security boundaries so enterprises can control agents’ knowledge scope, access rights, and behavior. This addresses enterprise concerns about controllability and security rather than simply providing models. Physical AI covers new customers in AI-RAN, robotaxis, and robots, extending the platform to the edge and the physical world. On the personal side, RTX Spark, after cooperation with Microsoft, brings similar capabilities down to Windows PCs, forming a continuous stack from the data center to the desktop.

Compared with other cloud or chip companies, NVIDIA emphasizes factory-level delivery and full-stack control (the five-layer model from energy to applications), rather than merely selling accelerators. The common point is competition for Europe’s sovereign AI buildout; the difference lies in modular pre-integration and an openly referenced agent security architecture. On the capital path, receptions from European governments and partner networks support the local deployment narrative, helping maintain share in discussions on data sovereignty and supply chains.

This statement changes expectations for AI infrastructure delivery: from on-site assembly to factory-prefabricated, plug-and-play modules. Metrics to watch are whether DSX POD begins shipping in early 2027, whether onsite deployment actually shortens to about 14 days, and how quickly OpenShell and Agent Toolkit are adopted in enterprise deployments; if European projects land on this schedule, NVIDIA’s reference position in regional AI factories will be strengthened.

ABAB News · Cognition Laws

  1. AI factories are shifting from on-site assembly to factory-prefabricated modules, compressing deployment time to the two-week range after the facility is ready.

  2. Agents need controllable boundaries: Toolkit integrates components, and OpenShell limits what they know, can access, and can do.

  3. Europe’s buildout path spans factories, agents, and physical AI, while RTX Spark on the personal side forms a continuous stack from cloud to desktop.