NVIDIA promises AI data centres using up to 30 percent less power and almost no cooling water

NVIDIA says its DSX MaxLPS software can let an AI data centre produce the same AI output with up to 30 percent less power. The company also says new cooling designs can cut water used for cooling to near zero. European cloud providers, energy companies and industrial software firms are adopting the platform, according to NVIDIA.

The news came at GTC Berlin on Wednesday 21 October 2026. NVIDIA Newsroom posted it on X at 11:42 Oslo time, the same day the company published a press release. Pricing and availability for the software are not disclosed.

What NVIDIA is launching

DSX is NVIDIA's reference architecture for what the company calls AI factories, data centres built to train and run AI models. Compute, networking, storage, software, power and cooling are designed as one system. The platform has several parts. MaxLPS allocates power dynamically within a data centre's provisioned power budget. Flex is meant to adjust power consumption to grid conditions. Sim builds digital twins of a facility before it is built, and DSX OS is meant to automate operations and fault-finding.

The partner list is long. NVIDIA names Schneider Electric, Siemens, Siemens Energy, ABB, Eaton, Vertiv, Trane Technologies, Hitachi Energy, GE Vernova, Dassault Systèmes and AVEVA, among others. Cloud providers CoreWeave, Google Cloud, Lambda, Nebius, Nscale and others are exploring MaxLPS, according to the press release.

The numbers NVIDIA presents

The key figure concerns power. NVIDIA says MaxLPS can deliver the same number of tokens with up to 30 percent less power, or more capacity within the same power budget. The figure is the company's own, and the press release does not describe the conditions under which it was measured.

Two partners have presented their own results. Nscale tested MaxLPS in its data centre in Iceland, on four racks of NVIDIA GB300 NVL72. Within the same 264-kilowatt power budget, the company saw 49 percent higher inference throughput and 37 percent more managed GPU capacity. Lambda reported 24 percent more throughput and 19 percent more GPU capacity in a proof of concept on HGX B200 systems. Both are evaluations, not sustained operations.

For Flex, NVIDIA points to an early trial with Emerald AI, the research institute EPRI, Nebius and the UK National Grid. A GPU cluster's power draw was cut by up to 40 percent in response to simulated grid events, while critical workloads kept running.

On cooling, NVIDIA builds on closed-loop liquid cooling and dry coolers that do not evaporate water. The Vera Rubin platform supports coolant temperatures of up to 45 degrees Celsius. According to NVIDIA, a 2-gigawatt facility can save 54 million litres of water a day compared with conventional cooling towers. The company itself notes that local climate determines whether extra mechanical cooling is needed in the hottest periods.

What is not disclosed

The press release does not say what MaxLPS or Flex cost, whether the software is sold separately or bundled with hardware, or when it is generally available. Three new DSX OS components are in developer preview. It says nothing about Norway or Nordic data centres beyond Nscale's facility in Iceland.

What Norwegian leaders should note

• Power and grid connection are the bottleneck for new data centres in Norway. If software gets more out of the same power budget, it changes the maths on capacity and cost. Ask suppliers to document the gain on your own workloads, not just cite NVIDIA's figures.
• Flexible power consumption could become an argument in talks with grid operators about faster connection. That assumes the business can tolerate some AI work slowing down when the grid is under strain.
• Customers buying AI capacity in the cloud should ask about power and water use per workload. That is becoming more important for sustainability reporting and procurement requirements.

Sources and media

• NVIDIA Newsroom on X: European players adopt NVIDIA DSX, post published 21.10.2026 at 11:42 Oslo time
• NVIDIA: European Ecosystem Builds World’s Most Efficient AI Factories With NVIDIA DSX, press release dated 21.10.2026
• Sources were checked on 21.10.2026.
• Illustration: AI-generated with Grok Bot / hogby.ai