NVIDIA targets faster, greener AI buildouts in Europe

NVIDIA rolls out DSX POD at GTC Berlin

NVIDIA (NVDA) introduced DSX POD at GTC Berlin on Oct. 21, 2026 as a factory-built reference design for Vera Rubin AI factories. The system is produced by partners and packaged inside modular containers that are fully integrated and validated before leaving the manufacturing site. This approach shifts much of the traditional on-site build and test work into the factory, with the goal of making advanced AI infrastructure faster and simpler to deploy.

Each DSX POD is described as a shipping-container-sized enclosure that houses seven Vera Rubin NVL72 racks. The units arrive with rack-to-rack cabling, wiring and piping already completed, and with liquid cooling installed at the factory. NVIDIA (NVDA) says this pre-integration allows customers to focus on preparing power, cooling interfaces and networking at the destination site rather than building the AI cluster from scratch.

Once delivered to a prepared location, DSX PODs are designed to connect to existing power, cooling and networking infrastructure and then complete final validation. NVIDIA states that this can reduce the time from shipment to production from several months to as little as 14 days. Partner-built DSX POD AI factories are planned for delivery worldwide starting in early 2027.

Beyond the containerized format, NVIDIA is positioning DSX reference designs as a way to improve the resource efficiency of AI data centers. The platform supports 100% closed-loop liquid cooling and dry-cooler configurations, which are designed to minimize dependence on traditional cooling towers. NVIDIA says this can bring facility cooling water consumption close to zero compared with conventional approaches.

The company also states that DSX-based designs can use up to 30% less power for the same AI output. At a 2-gigawatt deployment scale, NVIDIA estimates that the combination of closed-loop liquid cooling and dry coolers could save about 54 million liters of water per day versus typical cooling-tower-based systems. These claimed savings are framed as part of building what NVIDIA calls highly efficient AI factories.

The broader DSX platform spans hardware, software and planning tools intended to streamline design and operation of AI facilities. Components include DSX MaxLPS for dynamic power allocation, DSX Flex to coordinate workload power consumption with grid conditions, DSX Reference Design, DSX Sim, DSX OS and DSX Exchange. Together they are meant to manage power, cooling and workflow orchestration around Vera Rubin-based infrastructure.

NVIDIA is working with a wide group of partners to deliver DSX POD AI factories and related systems. Named collaborators include Anduril, Armada, Bull, Crusoe, DayOne, Dell Technologies (DELL), Delta, Eaton (ETN), Foxconn (Ingrasys), GIGABYTE, HPE, Lenovo, Pegatron, Quanta Cloud Technology, Schneider Electric (SUp), Vertiv (VRT) and Wistron. These companies are expected to integrate DSX designs into their offerings and help bring containerized AI factories to customers starting in 2027.

Alongside DSX POD, NVIDIA announced the OpenShell Reference Architecture on Oct. 21, 2026. OpenShell is described as a full-stack design for building, containing, observing and enforcing safe AI agents. It combines NVIDIA Vera Rubin infrastructure for model inference, NVIDIA Vera CPUs for running agent harnesses and tools, and NVIDIA BlueField-4 DPUs running NVIDIA Sentry for out-of-band monitoring and in-silicon policy enforcement.

In promoting OpenShell, NVIDIA states that Vera Rubin NVL72 systems can deliver 30x higher throughput per megawatt for agentic workloads. The OpenShell Reference Architecture is available immediately, giving organizations a blueprint to pair high-density, power-efficient compute with added observability and enforcement layers. Together with DSX POD and the DSX platform, it rounds out NVIDIA’s push to standardize both the physical and logical foundations of large-scale AI deployments.