Nvidia wants to cut data center water use, but that’s not the same as fixing AI’s water problem

2026-06-22 · Source: AI News & Artificial Intelligence | TechCrunch · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Cloud Computing & IT Infrastructure, Environmental Science & Earth Systems · Depth: Intermediate, short

Summary

Nvidia has introduced a warm-water cooling system designed to significantly reduce data center water usage, with an executive claiming it can eliminate "pretty much all water usage" inside the facility. This closed-loop system circulates coolant, entering racks at 45˚ C (113˚ F) and exiting at 55˚ C (131˚ F), enabling passive radiators in many climates and potentially achieving a 100% reduction in on-site water consumption. However, this claim is based on a narrow definition of water use, excluding the substantial water footprint from electricity generation and chip manufacturing. External water consumption can double or triple a data center's total water footprint, meaning Nvidia's solution addresses only about a quarter to a third of the overall problem. With fossil fuel power plants, such as natural gas (1.17 L/kWh) and coal (2.2 L/kWh), projected to supply over 40% of new data center electricity through 2030, the broader water consumption challenge remains significant despite internal cooling efficiencies.

Key takeaway

For AI Architects evaluating data center infrastructure for sustainability, recognize that on-site water efficiency is only part of the equation. Your choice of electricity source profoundly impacts the true water footprint, often doubling or tripling it. Prioritize data centers powered by renewables like wind or solar, which use minimal water, over those relying on fossil fuels. This holistic view ensures you address the full environmental impact of your AI operations.

Key insights

Nvidia's new cooling system addresses only a fraction of AI data centers' total water footprint by ignoring external electricity generation.

Principles

• Data center water footprint extends beyond facility walls.
• Power source choice critically impacts water consumption.
• Closed-loop cooling reduces on-site water use.

Method

Nvidia's system pumps 45˚ C coolant through racks, exiting at 55˚ C, allowing passive radiators to dissipate heat without evaporative cooling or fans, thus eliminating on-site water use.

In practice

• Evaluate total water footprint, including energy.
• Prioritize renewable energy sources.
• Implement warm-water cooling for on-site savings.

Topics

• NVIDIA
• Data Center Cooling
• Water Footprint
• AI Infrastructure
• Renewable Energy
• Sustainability Metrics

Best for: CTO, VP of Engineering/Data, Executive, Director of AI/ML, AI Architect, Policy Maker

Related on AIssential

• Do we really need gigantic, noisy, water-guzzling datacentres ruining our communities? In this economy? | First Dog on the Moon — Large data centers pose significant environmental and community burdens for potentially trivial AI applications.
• World’s first wind-powered underwater datacentre starts operating in China — Submerged, wind-powered datacenters significantly cut energy and water use through natural cooling and renewables.
• Meta’s natural gas binge could power South Dakota — Large-scale AI data centers are driving significant increases in energy demand, challenging corporate climate pledges.
• ChatGPT's Real Water Use, India's AI Boom & What Hyperscalers Can't Do | Ft. Sify CEO — Data center design in India prioritizes water efficiency and optimized redundancy while addressing local market demands.
• How AI data centres cut annual water consumption by 50% — Liquid cooling for AI data centers can eliminate water consumption, addressing environmental concerns for high-density compute.
• How some data center operators are tackling their water use problems — Data center water use is a complex, localized issue with inherent trade-offs between water consumption and energy efficiency.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by AI News & Artificial Intelligence | TechCrunch.