Nvidia Announces DSX Pod to Shrink AI Data-Center Build Time to Two Weeks

During the GTC Berlin conference, Nvidia revealed a new modular solution called the DSX Pod, designed to deliver a complete AI-focused data-center inside a single shipping container. The pod houses seven Vera Rubin NVL72 racks, networking equipment, cabling and liquid-cooling infrastructure that are fully assembled and validated before they leave the factory. Once the container arrives at a prepared site, the only tasks required are connecting power, cooling and network links, a process Nvidia says can be finished in as little as 14 days compared with the typical six-month schedule for conventional builds.

• Build time reduced from six months to roughly two weeks.
• Closed-loop cooling eliminates evaporative water use, potentially saving 100 % of cooling water.
• Global shipments expected to begin in early 2027.

Why a Containerized Approach Matters for AI Scale-Out

Ian Buck, Nvidia’s vice-president for hyperscale and high-performance computing, described the DSX Pod as a “pre-assembled, pre-validated building block for AI factories.” The company has partnered with 17 ecosystem members—including Bull, Dell Technologies, HPE, Lenovo, Eaton, Schneider Electric and Vertiv—to produce the pods at scale. Buck emphasized that the modular format is intended to accelerate AI infrastructure rollouts not only in Europe, where the announcement was made, but worldwide.

Energy-Saving Cooling and Power Management

The pod’s cooling system combines liquid-cooled compute with external dry-coolers, creating a closed-loop that rejects heat to the air without evaporation. Nvidia claims this design can cut water consumption for cooling by up to 100 % and lower power draw by as much as 30 % for the same compute output. At a 2 GW deployment level, the company estimates a daily saving of 54 million liters of water compared with traditional cooling-tower solutions. Additional features such as DSX MaxLPS dynamically allocate power within a data-center’s budget, while DSX Flex enables workloads to adjust consumption in response to grid conditions.

Built-In Governance and Security Layer

Alongside the hardware, Nvidia introduced the OpenShell Reference Architecture, a software framework that monitors AI agents, enforces policy rules and can isolate or disconnect misbehaving workloads. BlueField DPUs act as dedicated security monitors between the compute racks and the agent harness. Nvidia cites early adopters of OpenShell—including Anthropic, AWS, Google Cloud, Meta, Microsoft, OpenAI and Perplexity—as examples of organizations strengthening AI-agent governance.

Target Use Cases and Market Impact

IDC analyst Ashish Nadkarni highlighted three scenarios where the DSX Pod could be especially attractive: sovereign AI installations that require physical isolation, compact deployments built for specific workloads, and edge locations needing rapid AI capability. By delivering a turnkey solution, Nvidia aims to help operators start generating revenue from AI models sooner, reducing the lag between hardware acquisition and model inference.

Roadmap and Future Capacity Plans

The first DSX Pods are slated for shipment in early 2027. Nvidia also announced a broader manufacturing expansion, planning to increase substrate capacity by 2.5 times over the next three years, boost advanced-packaging capability by more than 300 % by 2028, and raise optical-engine production 15-fold by mid-2027, with a further doubling by the end of 2028. These investments are intended to keep pace with growing demand for AI infrastructure and to support the DSX platform’s scaling ambitions.