NVIDIA CEO Jensen Huang Unveils AI Factory and Open Models at GTC Berlin

The Tempodrom in Berlin, a venue with a rich history dating back to its founding near the Berlin Wall in 1980, has been transformed into a hub for cutting-edge technology. Today, it hosts the NVIDIA GPU Technology Conference (GTC) Berlin, where attendees are gathered for the main event: Jensen Huang’s keynote address.

Huang, founder and CEO of NVIDIA, took to the stage with a question on his mind: What do L’Oréal, Revolut, Spotify, Merck, Vodafone, Zara, Porsche, Nestlé, and HSBC have in common? The answer lies not just in their European roots but also in their collaboration with NVIDIA on AI projects.

This keynote comes after a weeklong road trip across Europe for Huang, where he met with various companies and announced that NVIDIA is committing nearly $10 billion to the technology ecosystem across the continent. This investment will be used to support 12,000+ partners working alongside NVIDIA in this space.

Huang framed AI as a five-layer cake, explaining how energy powers advanced computers that are part of an AI factory. The factory generates tokens and digital intelligence from models running on it, which are then integrated into every application. He emphasized the importance of having infrastructure in place before developing AI models.

‘Long before the AI models emerge, the infrastructure has to exist,’ Huang said, drawing parallels with how the internet required networking and electricity needed a grid. This analogy highlights the need for foundational elements in building an AI ecosystem.

To be part of this AI economy, companies don’t just want to consume AI; they also want to produce it. Huang stated that NVIDIA has already started working on enabling Europe to participate fully in the AI ecosystem. He emphasized the potential for European companies to not only use but also create their own AI solutions.

Huang presented NVIDIA’s full-stack AI factory as a comprehensive platform for building and operating AI infrastructure. This includes hardware, system software, CUDA-X libraries, and models – all of which are necessary for creating an AI factory. He framed this as the foundation upon which Europe can become part of the global AI economy.

He introduced Vera Rubin, in full production, as a significant extension of NVIDIA’s computing platform. It goes beyond just processing large language models to handle the entire AI agent – including its reasoning and decision-making capabilities. Huang also showcased several of NVIDIA’s latest rack-scale hardware offerings during his presentation.

Next, Huang announced NVIDIA DSX POD, a reference design for complete NVIDIA Vera Rubin AI factories built by partners inside modular containers. This innovation enables onsite deployment to shrink from six months to just 14 days once site utilities are ready – marking a significant transformation in the world supply chain.

‘This is a complete transformation of the world supply chain,’ Huang said. ‘This will change the way that people install and build data centers, and hopefully, as a result, velocity goes up.’ The NVIDIA DSX AI Factory Platform also offers benefits such as reducing water consumption for cooling by up to 100% and generating the same AI output with up to 30% less power.

Huang discussed agents in detail, explaining that they think, reason, iterate, and use tools – driving token consumption as they become more useful. He framed an agent as comprising five elements: a model, a harness for orchestrating its work and memory, a sandbox isolating the harness, knowledge, and tools.

The most critical aspect of agents is ensuring their security. Huang emphasized that since autonomous agents can access files, resources, networks, and tools, their privileges should be minimal and they must be monitored closely. NVIDIA’s response to this challenge is the new NVIDIA OpenShell Reference Architecture for Safe AI Agents – a full-stack design built on NVIDIA OpenShell.

Open Models Turn Expertise Into an Advantage

Huang presented NVIDIA’s open models as starting points for building AI tailored to specific needs. These frontier, open models are available across industries: Nemotron for language, Cosmos for world foundation models, GR00T for humanoid robotics, Alpamayo for autonomous vehicles, Ising for quantum-classical computing, and Earth-2 for physics AI.

NVIDIA provides the models, data, and scripts to reproduce them. Companies can post-train a model on their own data for a particular use case – turning proprietary data into what Huang called ‘sovereign intelligence.’ This approach allows companies to start with an open model like Nemotron and achieve superintelligence by adapting it to their specific needs.

NVIDIA Gives Agents a Knowledge Layer

Huang introduced the NVIDIA Agent Knowledge Foundation as a set of models and libraries for connecting agents to enterprise data. He highlighted Nemotron tools for parsing documents, embedding queries, and reranking relevant information – along with new NVIDIA Kumo models for making predictions from structured data.

‘Ultimately, what turns these super agents into super agents is that you give them access to your company’s knowledge,’ Huang said.