Nvidia has recently introduced an innovative closed-loop, warm-water cooling system for its data centers. This technology is designed to operate at significantly higher temperatures than traditional systems, with water pumping in at 45°C (113°F) and out at 55°C (131°F).
By allowing heat to be removed via passive radiators rather than relying on energy-intensive evaporative cooling, Nvidia claims this system can eliminate almost all water usage inside the data center. While this sounds like a massive leap forward for environmental sustainability in the AI sector, a closer look reveals that it might not be the complete solution it appears to be.
## The “Boundary” Problem
The primary critique of Nvidia’s new approach is that it focuses exclusively on water consumption within the physical walls of the data center. This is often referred to as the “boundary” problem.
By measuring efficiency solely based on direct water usage for cooling the servers, the company effectively ignores the massive indirect water footprint associated with the broader AI infrastructure.
## The Hidden Water Footprint: Electricity and Chip Manufacturing
The water used directly for cooling is only a fraction of the total water required to sustain an AI data center. Water use outside the data center—specifically for electricity generation and chip manufacturing—can be two to three times greater than the cooling consumption of the facility itself.
Power plants, especially those relying on fossil fuels, consume vast amounts of water for cooling their own systems. Furthermore, the semiconductor manufacturing process, which creates the very chips that power Nvidia’s AI empires, is notoriously water-intensive.
| Water Use Type | Description | Proportion of Total Footprint |
| :— | :— | :— |
| **Direct (Data Center Cooling)** | Water used on-site to cool server racks. | ~25% – 33% |
| **Indirect (Electricity Generation)** | Water used by power plants to generate the electricity that runs the data center. | ~33% – 50% |
| **Indirect (Chip Manufacturing)** | Water used in the fabrication of silicon chips and hardware. | ~25% – 33% |
## The Fossil Fuel Connection
The indirect water footprint is further exacerbated by the tech industry’s growing reliance on fossil fuels. As the energy demands for training and running massive AI models skyrocket, tech companies are increasingly turning to natural gas plants, sometimes even building their own, to ensure a stable power supply.
The water required to support this energy generation remains a critical, and largely unaddressed, component of AI’s total environmental impact. Until the tech industry can fully transition to renewable energy sources that do not rely on water-intensive cooling, the AI water problem will persist, extending far beyond the walls of any single data center.
