Back to the MIT repository
6. Socioeconomic and Environmental1 - Pre-deployment

Environmental cost (water consumption)

Data centers use water for cooling to prevent servers from overheating. The water consumption associated with AI training and inference processes can be substantial, impacting local water resources.

Source: MIT AI Risk Repositorymit754

ENTITY

3 - Other

INTENT

2 - Unintentional

TIMING

1 - Pre-deployment

Risk ID

mit754

Domain lineage

6. Socioeconomic and Environmental

262 mapped risks

6.6 > Environmental harm

Mitigation strategy

1. Prioritize the deployment of advanced, low-water-intensity cooling architectures, such as liquid immersion and direct-to-chip cooling, to substantially reduce reliance on water-intensive evaporative cooling systems. 2. Implement comprehensive, transparent water accounting protocols to rigorously measure and publicly disclose the total water footprint—including Scope 1 (operational), Scope 2 (energy), and Scope 3 (embodied) consumption—to enable water-aware decision-making. 3. Strategically site new data centers in non-water-scarce regions, and for existing facilities, mandate circular water management practices, including wastewater reclamation and local watershed replenishment, to mitigate impact on municipal supplies.