Nvidia announced a new closed-loop liquid-cooling system this month designed to reduce the on-site water consumption of AI data centers to near-zero [1, 3].

The technology arrives as the environmental cost of artificial intelligence draws scrutiny. Massive computing clusters require immense cooling to prevent hardware failure, often relying on freshwater that strains local ecosystems and municipal supplies.

Recent disclosures highlight the scale of the issue. Amazon said its data centers used nine billion liters of water [1]. The consumption occurs not only at the facility level but also through the energy grid. A 2025 peer-reviewed study indicated that generating a 100-word email consumes approximately 519 ml of water [2].

Nvidia's new design utilizes a closed-loop system to maintain performance while eliminating the need for constant water replenishment [1, 3]. This approach targets the immediate water footprint of the hardware racks. However, some analysts said the fix is limited. TechCrunch reported that while the system cuts water use inside the data center, it does not address the water required by the fossil-fuel power plants that provide the electricity [3].

Regional responses to AI expansion vary. In Scotland, some water authorities said the country's water resources could accommodate additional AI data-center capacity [4]. This contrasts with broader reports of AI facilities draining freshwater reserves in other parts of the world [2].

The shift toward liquid cooling reflects a broader industry trend to move away from traditional evaporative cooling. By circulating coolant in a sealed environment, operators can avoid the massive evaporation losses associated with cooling towers, a primary driver of high water usage in traditional cloud infrastructure [1, 3].

“Nvidia's new design utilizes a closed-loop system to maintain performance while eliminating the need for constant water replenishment.”

While closed-loop cooling mitigates the direct local impact of data centers on municipal water tables, it shifts the environmental focus to the energy grid. The total water footprint of AI remains a combination of direct facility cooling and indirect water use at power generation sites, meaning hardware efficiency alone cannot solve the industry's sustainability crisis.