NVIDIA opened its GTC Berlin 2026 trade show on Tuesday with a keynote from CEO Jensen Huang that put the company's next-generation Vera Rubin platform at the center of its European push — and, in a literal twist, inside a shipping container. The headline announcement, the DSX POD, is a reference design that packs a complete multi-rack AI data center into a standard intermodal container that can be shipped by truck and installed in about two weeks, according to ServeTheHome's coverage of the keynote.
The show, NVIDIA's principal trade event for Europe, comes as the company leans harder into a market it has repeatedly described as underserved. Associated Press reporting from Berlin framed the keynote as an effort to encourage Europe's lagging AI sector with tools to accelerate data center buildout, and NVIDIA's own newsroom pointed to new European public sector projects built on its open models. For more context on this story, see our ongoing breaking AI news.
What the DSX POD Actually Is
The DSX POD is not to be confused with NVIDIA's DGX SuperPOD, the company's blueprint for large cluster builds. Instead, the DSX POD is far more literal: a fully assembled, tested, and validated computing system built inside a standard shipping container, most likely a 40-foot unit, though NVIDIA has not confirmed the exact size.
According to ServeTheHome's technical breakdown, the reference design calls for seven Vera Rubin NVL72 racks inside the container, along with the associated networking, cooling, and management hardware. The result delivers roughly 25 exaflops of AI inference per container. The units are not fully self-contained — they still need power, network connectivity, and a larger cooling infrastructure — but beyond utilities, each POD is a complete, working AI system on its own.
The point of the design is speed and repeatability. Because the containers are assembled and validated at the factory, a standardized DSX POD can be delivered and installed within 14 days rather than the months a traditional on-site build requires. And because the design uses standard intermodal container dimensions, the units can ship by truck rather than heavy rail, which means they can reach sites that conventional data center construction struggles to serve.
NVIDIA will not build or sell the PODs itself. The partner list reads like a cross-section of the server industry — Gigabyte, HPE, Dell, Delta, and Quanta Cloud Technology (QCT) among them — and NVIDIA expects partners to begin delivering DSX PODs worldwide in early 2027.
OpenShell: A Blueprint for Keeping AI Agents in Line
The most consequential software announcement may be OpenShell, a reference architecture for agentic AI safety. As enterprises move from chatbots to autonomous agents that execute code, call tools, and act on live systems, the industry has struggled with how to contain the blast radius when an agent misbehaves — a concern underscored by a string of recent security incidents where AI systems were pushed to their limits.
OpenShell is NVIDIA's answer: not a new chip or a standalone product, but an architectural guide for building a secure operating environment around agentic AI workloads. The reference architecture pairs the Vera Rubin platform for large model reasoning with NVIDIA's Vera CPUs handling agent execution and the OpenShell secure runtime stack, while BlueField-4 DPUs act as the hardware-level security enforcement layer. The documentation and supporting software became available at the show.
The approach reflects a broader hardware trend: as agent deployments scale, security is migrating from software-only controls into dedicated silicon on the network path. Yahoo Finance's coverage of the keynote similarly highlighted NVIDIA's platform for keeping AI agents from going rogue as one of the show's marquee items.
Supply Chain Promises: Substrates to Optics
Huang also used the keynote to detail how NVIDIA intends to scale its supply chain over the next three years, a critical question for an industry that has spent the past half-decade bumping against capacity ceilings in wafers, memory, packaging, and testing. Per ServeTheHome, the targets NVIDIA laid out include:
- Substrate capacity up 2.5x over three years, sourced from IBIDEN, Samsung Electro-Mechanics, and Unimiron
- A 4x increase in interposer wafers for advanced packaging from Amkor by the end of 2028
- A doubling of test capacity at KYEC by 2027
- A 15x increase in optical engine capacity for co-packaged optics from SPIL, doubling again to 30x by the end of 2028
The optics number is the standout. NVIDIA is moving aggressively into co-packaged optics with the Vera Rubin generation, and the jump from small-volume production to a 30x capacity expansion signals how central optical interconnects have become to the company's roadmap for dense, power-efficient AI networking.
Europe's Moment, With a Protest Chorus
The Berlin setting was no accident. Europe is preparing a wave of sovereign AI investments and supercomputer builds, and NVIDIA wants its hardware at the center of them. Reuters described Huang appearing flanked by robot ducks as he hailed Europe's AI potential, while NVIDIA's newsroom highlighted European financial services firms embracing open models to build proprietary AI systems on its platform.
The keynote did not go entirely to plan. A Greenpeace protester disrupted the speech, demanding answers over the use of chips in Russian drones before being removed, according to Tom's Hardware's report from the show — a reminder that AI infrastructure's geopolitical footprint is drawing activism alongside investment.
For smaller-scale operators, there was news too: an update to NVIDIA's Sync Cluster Assistant now lets DGX Spark units cluster up to eight devices, a notable bump for local AI labs built on the compact systems.
Why It Matters
The DSX POD crystallizes where data center construction is heading: away from bespoke, months-long construction projects and toward standardized, factory-built modules that arrive like appliances. If containerized AI factories deliver on the promised 14-day installation timeline, the pace at which new compute capacity can come online — particularly in regions like Europe playing catch-up — could accelerate sharply.
Combined with the OpenShell safety architecture and the supply chain commitments, this year's GTC Berlin keynote sketched out NVIDIA's full-stack bet: more capacity, delivered faster, with hardware-enforced guardrails around the agents running on it.
Stay Ahead of AIFollow AI Buzz Wire for breaking AI news, model releases, and industry analysis.
Read more AI news →
