Nvidia data center water use slashed: New cooling tech challenges AI’s thirsty footprint

Nvidia data center water use: a breakthrough in cooling

As AI workloads balloon, the thirst of data centers has sparked backlash from communities and regulators. Nvidia claims its Vera Rubin servers can cut onsite water consumption to near‑zero by running coolant at 45 °C and allowing it to rise to 55 °C before dry‑cooler rejection.

Key innovation: a closed‑loop fluid of water and propylene glycol that tolerates higher temperatures, eliminating the need for evaporative towers in most climates.

“The 45‑degree intake temperature is transformative,” says Josh Parker, Nvidia’s head of sustainability.

Traditional air‑cooled farms rely on evaporative cooling, a process that can demand ↓5 million gallons of water daily for a large facility—comparable to a small town.

Liquid cooling reduces the share of power devoted to climate control to ↑10% of total draw, according to Nvidia’s data.

Nevertheless, indirect water use persists via electricity generation. A 2024 DOE study cited by Reuters estimates AI‑related power consumption generated 211 billion gallons of water in 2023, enough for New York City’s seven‑month supply.

Experts warn that while Nvidia’s system eases onsite demand, peak‑load planning and grid mix will keep water footprints sizable. Shaolei Ren of UC Riverside notes that utilities must accommodate occasional spikes when evaporative cooling is still needed.

In practice, operators can shift workloads to cooler regions or defer non‑critical jobs on scorching days—tactics reminiscent of load‑shedding strategies during the pandemic surge.

Bottom line: Nvidia’s hotter‑running coolant offers a meaningful reduction in direct water use, but it does not resolve the broader resource debate surrounding AI’s expansion.

Intel provided by: Dr. Aris Thorne

Artificial Intelligence Researcher