# NVIDIA Targets Near-Zero Cooling Water Use at European AI Factories By Simon Yoon Canonical URL: https://www.tokenpost.com/news/technology/37377 Published: 2026-10-22T03:10:28.000Z Updated: 2026-10-22T03:10:28.000Z Section: Technology > At a 2-gigawatt scale, the design is projected to save 54 million liters of water a day. MaxLPS is designed to produce the same number of tokens with up to 30% less power. NVIDIA announced Oct. 21 at GTC Berlin that European partners are adopting its DSX AI Factory Platform, whose cooling design is projected to save 54 million liters of water daily at a 2-gigawatt scale. DSX is a reference architecture that combines computing, networking, storage, software, power and cooling. Its cooling system circulates liquid coolant through computing equipment, while dry coolers release heat into the air without evaporating water. Local climate conditions may require supplemental mechanical cooling during the hottest periods. At a 2-gigawatt deployment scale, the design is projected to save 54 million liters of water per day compared with conventional cooling towers, roughly the daily water use of 180,000 households. That figure is a projection, not a measurement from an operating facility of that size. The near-zero water claim concerns on-site cooling. U.S. data centers directly used an estimated 66 billion liters of water in 2023, while nearly 800 billion liters of indirect water consumption was associated with the electricity they used. The figures describe different parts of the data center water footprint. DSX also includes MaxLPS, software for allocating power across an AI factory. It is designed to produce the same number of tokens with up to 30% less power. A separate Lambda proof of concept on HGX B200 systems showed cluster token throughput rising 24% and GPU capacity increasing 19%, with power held constant. A joint evaluation at Nscale’s Iceland facility used four GB300 NVL72 racks. The number of managed GPUs rose from 140 to 192, while aggregate inference throughput increased from 1,084,503 to 1,618,443 tokens per second under the same 264.4-kilowatt power budget. The evaluation also recorded a 17% increase in 99th-percentile time to first token, a response-time measure, from a 15.7-second baseline. CoreWeave, Firmus, Google Cloud, IREN, Lambda, Nebius, Nscale and Sharon AI are exploring MaxLPS. Nscale founder and CEO Josh Payne said, “Europe’s AI ambitions demand that we make every megawatt count.” The cooling effort is part of NVIDIA’s broader factory push, which includes [DSX POD modular systems]() intended to shorten deployment at prepared sites. ## Links in this article - [DSX POD modular systems](https://www.tokenpost.com/news/technology/36963)