GTC Berlin 2026: Live Updates From NVIDIA CEO Jensen Huang’s Keynote

At GTC Berlin, NVIDIA's chief executive framed Europe as ready to produce AI rather than consume it, capping a week in which the company says it committed nearly $10 billion to Europe's technology ecosystem and works with more than 12,000 partners. The keynote ran the stack top to bottom: Vera Rubin in full production, a modular DSX POD reference design, open models, an agent knowledge layer, telecom, healthcare, cars, a reference humanoid and RTX Spark laptops.

The headline for us is how much of it was about agents and how to keep them contained. When the world's dominant AI hardware company says the most important thing is making sure the agent is secure, that sets the agenda for every enterprise buyer listening.

We're encouraged by the security framing, and not just as stagecraft. Huang described an agent as five parts: a model, a harness that orchestrates work and memory, a sandbox that isolates the harness, knowledge and tools. He argued agent privileges should be minimal, monitored and adjusted every time an agent launches. That's the right mental model. The answer NVIDIA is selling is the OpenShell Reference Architecture, an open-source, policy-controlled runtime with kernel-level sandboxing between the agent and the systems it can affect. Whether it works in the wild is the test, but naming least privilege as the default is a step the industry needed.

The infrastructure news is the most consequential for Europe. NVIDIA says DSX POD, prebuilt Vera Rubin AI factories in modular containers, can shrink onsite deployment from six months to 14 days once site utilities are ready. It also says its DSX AI Factory Platform can cut cooling water use by up to 100% and produce the same output with up to 30% less power. Those are vendor claims with caveats baked in, notably the utilities part. Power and grid access, not containers, are what slow data centers in Europe.

The sovereignty pitch deserves a skeptical ear. Huang urged companies to post-train open models like Nemotron on their own data and called the result "sovereign intelligence." Open models with reproducible data and scripts are genuinely useful, and we'd rather Europe build on open weights than rent closed ones. But sovereignty that runs on one American company's chips, libraries and reference designs is a narrower kind of independence than the word suggests.

The applied examples were the strongest moments. Agents at Lila Sciences helping design a potential B-cell lymphoma therapy, Siemens testing production plans in a digital twin, and adaptive radiation therapy with Siemens Healthineers that tracks a changing tumor during treatment. That last one is where useful AI and serious harm sit closest together. A system that adjusts radiation targeting in real time needs validation and regulatory scrutiny far beyond a keynote video.

Where this leads if it scales: every European enterprise running agents on NVIDIA's full stack, faster data center builds and more of the continent's compute tied to one vendor's roadmap. Some of that is progress. We'll keep asking who holds the keys.

The keynote took place at Berlin's Tempodrom, with GTC Berlin running three days and more than 100 sessions. Other announcements included the Agent Knowledge Foundation and Kumo models, a first commercial phone call on T-Mobile's network with Nokia, wider adoption of the Hyperion automotive platform by Volvo Cars and others, the Agile ONE SN reference humanoid expected next year, and RTX Spark laptops that launched the prior week.