NVIDIA DSX POD ships AI factories in containers

What changed?

On October 21 at GTC Berlin, NVIDIA announced DSX POD (NVIDIA Newsroom), a reference design for complete Vera Rubin AI factories that partners build inside modular containers and integrate and validate before delivery. Compute, networking, cabling, rack-level power and liquid cooling are all installed at the factory.

According to NVIDIA, deployment at a prepared site shrinks from six months to as little as 14 days. DSX POD units ship in early 2027, with partners including Dell, HPE, Lenovo, Foxconn, Schneider Electric, Vertiv and Crusoe.

More DSX updates

In a separate announcement (NVIDIA Newsroom), NVIDIA expanded the DSX software stack:

• DSX MaxLPS allocates power dynamically inside an AI factory. NVIDIA says it can produce the same tokens with up to 30% less power. Lambda measured 24% more token throughput on B200 systems within a fixed budget, and Nscale's Iceland test showed 49% higher inference throughput on four GB300 NVL72 racks.
• DSX Flex adjusts power draw to grid needs. In a U.K. trial, cluster power dropped by up to 40% during simulated grid events while critical workloads kept running.
• DSX OS adds three open source components in developer preview (Config Manager, Rack Management Service, Cluster Readiness Engine) to automate network setup, rack management and pre-production cluster validation.
• SimReady Central brings validated power and cooling equipment models to Hugging Face for faster digital twins.

NVIDIA says closed-loop liquid cooling with dry coolers can cut facility cooling water use to near zero, saving 54 million liters a day at 2-gigawatt scale.

Why it matters

The bottleneck for AI capacity is often construction, power and cooling rather than GPUs. DSX POD targets deployment time, while MaxLPS and Flex target the power budget. The figures are vendor or partner measurements without independent verification. For Europe's sovereign AI plans, faster and more predictable builds may be the real advantage.