Nvidia said at London Climate Week that its next generation of AI infrastructure can be cooled in a way that may significantly reduce data centers' water consumption. The company announced that its latest system can be fully cooled with a warm liquid, potentially reducing or removing the need for additional chilling equipment.
What did they say?
Josh Parker, Nvidia’s chief sustainability officer, told reporters in an interview before traveling to London: “The water consumption challenge for data centers is largely solved.” Nvidia described a recirculated coolant mixture that includes water and propylene glycol — similar to automotive antifreeze — which the company says can operate at 113 degrees Fahrenheit (about 45 °C).
Because the liquid can run at higher temperatures than previous systems, data centers may rely less on or eliminate mechanical chillers that consume large amounts of energy or water, thereby lowering cooling-related water and energy use.
Why it matters
The announcement comes as Google and Amazon have faced scrutiny and local opposition over data center water practices while AI infrastructure expands. Technology companies increasingly argue that efficiency gains will mitigate the environmental impacts of AI buildouts.
Industry reactions
Steve Solomon, Microsoft’s vice president of data center engineering, commented on the potential before Nvidia’s announcement was public, saying: “It would be a big deal for everybody if we got all of the chips to do that.” Solomon added that in most climates — even in hot places such as Arizona — the approach could remove the need for mechanical chillers most of the time.
Limits and open questions
Even if Nvidia’s technology substantially reduces cooling-related water use, that does not erase all water concerns. Widespread deployment of the new systems will take years, and many existing data centers will continue to use older cooling technologies. Nvidia declined to provide details on system costs; the pace of adoption will likely depend on the economics of facilities designed for fully liquid-cooled AI infrastructure, though Nvidia says operators will save on cooling costs.
Also, water use inside a data center is only part of the broader debate: generating the electricity to run AI infrastructure can require significant water depending on the power source.
What’s next
Nvidia’s approach could make each unit of AI computing more efficient, but the company is explicit that those efficiency gains are not intended to constrain growth. As Parker wrote in a blog post, “AI workloads are not getting lighter.” Without efficiency improvements, he argues, the energy required to run AI would keep rising with demand.
Observers will be watching whether efficiency improvements reduce per-unit water and energy use and whether those same gains accelerate the buildout of AI infrastructure, ultimately increasing the sector’s overall footprint.
