Artificial intelligence’s rapid scale-up is quietly driving emissions and water use toward city-scale levels—yet weak disclosure makes the true footprint far larger and murkier than reported.
Disclaimer: VoD Capsules are AI-generated. They synthesize publicly available evidence from reputable institutions (UN, World Bank, AfDB, OECD, academic work, and other such official data sources). Always consult the original reports and primary data for verification.
Executive Summary
This Patterns (2025) perspective by Alex de Vries-Gao delivers one of the clearest warnings yet: AI’s environmental footprint is accelerating faster than our ability to measure or govern it . Drawing on public sustainability reports from major tech firms (Google, Meta, Microsoft, Apple, Amazon, Tencent, Baidu), the study shows that no company meaningfully discloses AI-specific energy, carbon, or water data. As a result, analysts must infer AI impacts indirectly from overall data-center performance—an approach that systematically understates risk.
Using International Energy Agency baselines and company disclosures, the paper estimates that AI systems alone could emit 32.6–79.7 million tons of CO₂ in 2025, comparable to the annual emissions of New York City, while water use could reach 312–765 billion liters—approaching global bottled water consumption. Critically, indirect water use embedded in electricity generation is likely severely underestimated by official statistics.
These findings align with investigative reporting (e.g., The Guardian, Dec 2025) highlighting how the AI boom is straining grids and water systems, often in regions already facing scarcity. Together, they underscore a governance gap: AI is scaling like heavy industry, but regulated like software.
Think About It This Way
AI’s footprint isn’t just about clever algorithms—it’s about where servers are built, which grids power them, and whose water cools them. When disclosure is opaque, environmental costs don’t disappear; they simply become someone else’s problem.
Implications (What This Means in Practice)
- AI is becoming a city-scale infrastructure user
Emissions and water demand now rival those of major urban economies, shifting AI firmly into climate and resource policy territory. - Water risk is the blind spot
Indirect water use from electricity generation often exceeds on-site cooling, yet remains largely undisclosed and ungoverned. - Geography matters more than efficiency slogans
Identical AI workloads can have vastly different impacts depending on grid carbon intensity and local water stress. - Corporate averages obscure local harm
Company-wide sustainability metrics mask the concentrated burdens faced by specific communities hosting data centers. - Transparency is now a systems constraint
Without granular disclosure (AI vs. non-AI workloads, site-level PUE/WUE), neither markets nor policymakers can steer mitigation effectively.
Further Reading
| Report / Study | What it covers / Why useful | Official Link |
|---|---|---|
| de Vries-Gao (2025), Patterns | Core estimates of AI carbon & water footprints | https://doi.org/10.1016/j.patter.2025.101430 |
| IEA (2025), Energy and AI | Global data-center energy baselines | https://www.iea.org/reports/energy-and-ai |
| Masanet et al. (2024), Joule | Why better AI energy data is urgently needed | https://doi.org/10.1016/j.joule.2024.07.018 |
| LBNL (2024) | Grid-level carbon & water intensity mapping | https://waterimpacttool.lbl.gov/ |
| The Guardian (2025) | Investigative link between AI boom, CO₂, and water stress | https://www.theguardian.com/technology/2025/dec/18/2025-ai-boom-huge-co2-emissions-use-water-research-finds |
Explore Further
- Explore further: How would mandatory AI-specific disclosure (energy, water, location) change investment and siting decisions?
- Explore further: Which regions risk becoming “AI sacrifice zones” due to cheap power or weak water governance?
- Explore further: What would it take to regulate AI infrastructure more like energy or transport systems, not apps?
VoDGPT is an AI system powered by OpenAI, and it can make mistakes.
Use VoD Capsules as a starting point for understanding; always review the linked reports and verify critical information.
About the Author
