Public pushback against the massive water and energy consumption of AI data centers is forcing a hardware rethink, and Nvidia claims to have a radical solution. By switching entirely to liquid cooling and allowing servers to run significantly hotter, the company says its next-generation data center design can eliminate almost all water usage. The shift targets a growing environmental crisis as generative AI demands unprecedented computational power.
Nvidia is currently highlighting its claim that the Rubin generation reference design for a fully liquid-cooled data center has eliminated massive amounts of power usage. The efficiency gains are largely achieved by pushing the thermal limits of the hardware, allowing AI servers to run as hot as 113 degrees Fahrenheit (45 degrees Celsius). In this system, heat is captured directly at the chip and transported through liquid loops operating at much higher temperatures.
This high-temperature liquid approach allows outdoor dry coolers to reject heat efficiently for most of the year, offering greater flexibility regarding ambient air temperatures. The environmental impact of this architectural shift could be staggering for local municipalities that currently supply millions of gallons of water to tech facilities.
From roughly 2.6 million gallons per megawatt per year for conventional cooling-tower-based systems to near zero - up to a 100 percent reduction.
- Josh Parker, Head of Sustainability, Nvidia
However, the transition is not without its blind spots. As noted by tech outlet Gizmodo, Nvidia's announcement conveniently omits the massive capital expenditure required to build this style of data center compared to less efficient air-cooled facilities. Furthermore, the design does not address the immense power generation requirements needed to run these massive AI factories, nor the carbon footprint of their initial construction.
Nvidia is not alone in exploring higher thermal tolerances to improve efficiency. In a recent report, Amazon similarly touted higher heat tolerances as a key strategy for making its mostly air-cooled data centers more efficient. Despite the high costs, Nvidia claims that every cloud provider and data center operator building for the Rubin architecture is making the transition to liquid cooling.
The Hidden Cost of Running Hot
While saving 2.6 million gallons of water per megawatt is a massive public relations and environmental win, the transition to 100 percent liquid cooling requires a fundamental, expensive overhaul of existing global infrastructure. Cloud providers will have to weigh the upfront capital expenditure of retrofitting facilities against long-term operational savings and regulatory goodwill. The era of cheap, air-cooled server racks is effectively over for top-tier AI workloads.
Furthermore, running silicon at 113 degrees Fahrenheit pushes the thermal limits of surrounding hardware components. While the GPU itself might survive these blistering temperatures, the ambient heat degrades networking cables, storage drives, and power supplies much faster. Facility operators may find that the money they save on water bills is quickly reallocated to replacing failed hardware and managing complex liquid-loop maintenance.
