AI Companies' Water Disclosure Gap: Google Leads Transparency, But Industry Lacks Standardization
Key Takeaways
- ▸Google consumed 8.1 billion gallons of water in 2024, representing an 88% increase since 2019, driven by AI infrastructure expansion
- ▸Google is among the few AI companies publishing per-query water efficiency metrics and facility-level data, though disclosure remains incomplete across the industry
- ▸The AI industry lacks standardized water reporting frameworks, making it difficult to compare environmental impact across competitors and hold companies accountable
Summary
A new open dataset reveals significant disparities in how major AI companies disclose water consumption data, with Google emerging as the most transparent. Google reported consuming 8.1 billion gallons of water in 2024—an 88% increase since 2019—and published fleet-wide water usage effectiveness (WUE) metrics of approximately 1.1 liters per kilowatt-hour. The company also provided per-query water estimates for its Gemini model (0.26 ml) and disclosed a carbon-free energy percentage alongside facility-level data for some locations like Council Bluffs.
Despite these efforts, the dataset highlights inconsistencies across the AI industry in environmental reporting standards. While Google has committed to replenishing 120% of its water consumption by 2030 and achieved approximately 64% replenishment (4.5 billion gallons) in 2024, most competitors lack comparable transparency around facility-level water metrics and per-model efficiency data. The absence of systematic, standardized water disclosure across the industry makes it difficult to benchmark environmental impact and holds companies accountable for their growing computational footprint as AI workloads scale globally.
- Google's water replenishment pledge (120% by 2030) achieved 64% completion in 2024, demonstrating progress but revealing the scale of the water challenge ahead
Editorial Opinion
Transparency in water consumption is becoming as critical as carbon disclosure for AI companies operating at scale. Google's willingness to publish per-query water metrics and facility-level data sets a commendable precedent, yet the fragmented disclosure landscape suggests the industry is unlikely to self-regulate effectively. Without standardized reporting frameworks mandated by regulators or industry bodies, meaningful comparison and accountability remain elusive—and the true environmental cost of training and deploying large language models risks being obscured.
