Tech • AI • Robotics • Game
AI Factories That Use Less Power and Water, Built in Days | NVIDIA DSX
9.5/10TL;DR
NVIDIA has unveiled DSX and the factory-built DSX Pod, a full-stack AI factory architecture designed to cut deployment time, reduce power use, and sharply lower water consumption at hyperscale.
KEY POINTS
Full-stack AI factory design
DSX is positioned as a reference architecture that treats compute, networking, storage, software, power and cooling as a single co-designed system rather than separate layers. The approach aims to improve performance and operational efficiency as AI infrastructure expands into what is described as one of the largest industrial build-outs in history.
Digital twin before construction
The system is designed to begin in a digital twin, allowing performance, power, water use and scale to be validated virtually before any physical construction starts. That is intended to reduce design errors and speed decisions on facility layout, equipment integration and operating limits.
Broad industrial partnerships
PTC converts equipment designs into simulation-ready assets, while Dassault Systèmes links rack and facility design through model-based systems engineering. Schneider Electric simulates power distribution and liquid cooling, Jacobs integrates the overall digital twin in Omniverse, Siemens evaluates thermal, acoustic and grid effects, and Procore supports construction planning.
Power as the key production constraint
NVIDIA argues that power availability is the main factor limiting how much an AI factory can produce. Its DSX MaxLPS control loop dynamically allocates power to match workload demand and is claimed to use 30% less power for the same output.
Liquid cooling and water savings
Vera Rubin systems are designed for closed-loop liquid cooling with coolant temperatures of up to 45 degrees Celsius. In suitable climates, outdoor dry coolers can reject heat to the air without evaporating water, avoiding the heavy water use associated with traditional cooling towers.
Savings at hyperscale
At a facility scale of 2 gigawatts, DSX is projected to save 54 million liters of water per day. That volume is equated to the daily water use of 180,000 households, underscoring how cooling design is becoming a major infrastructure and sustainability issue for large AI sites.
Factory-built DSX Pod
The newly announced DSX Pod packages the platform into a factory-built, integrated system to shorten deployment. A Vera Rubin NVL72-based pod consists of seven liquid-cooled compute racks linked by 17 kilometers of Spectrum-X cabling.
Faster time to operation
DSX Pods are integrated, fully cabled and factory tested before being shipped in transportable containers. The design uses dry coolers instead of evaporative cooling and 800-volt DC power delivery to the racks, with on-site deployment expected to drop from several months to a matter of days once utilities are ready.
CONCLUSION
The DSX strategy reflects a shift in AI infrastructure from standalone servers to tightly engineered industrial systems where power, cooling and construction speed are as important as chips. With DSX Pod, the competitive edge increasingly depends on how quickly operators can turn capital and electricity into working AI capacity.
Ask a question