UN Warns AI Water Use May Exceed 1.3 Billion People's Drinking Needs
Abhishek Gautam12 min readQuick summary
UNU-INWEH: AI data centers could consume 9.3 trillion liters of water by 2030 and 945 TWh of power — carbon, water, and land footprints move in opposite directions.
Read next
- O'Leary Cuts 40,000-Acre Utah AI Data Center in Half After Backlash
- IEA Just Released 400 Million Barrels of Oil. It Did Not Work. Here Is What That Means for Tech.
A United Nations University report published June 4, 2026 warns that AI data centers could consume 9.3 trillion liters of water by 2030 — equivalent to the basic annual domestic water needs of all 1.3 billion people in Sub-Saharan Africa — while electricity use nearly doubles from 448 TWh (2025) to 945 TWh.
UN News and The Conversation summarized the finding bluntly: AI's resource footprint is not just carbon — and "low-carbon" energy can be high-water.
What UNU-INWEH Measured
Report title: "Environmental Cost of AI's Energy Use: Carbon, Water and Land Footprints" (UNU-INWEH).
| Footprint | 2025 baseline | 2030 projection |
|---|---|---|
| Electricity | 448 TWh (~France-scale) | 945 TWh (~3% of global power) |
| Water | (growth trajectory) | 9.3 trillion liters |
| Land | — | >14,500 km² (~2× Jakarta metro, ~10× Mexico City) |
| CO₂ equivalent | — | ~400 Mt (~UK annual emissions) |
Per-generation micro-costs (today):
- One AI image: ~29 mL water (electricity-associated); 17 minutes of a 10W LED
- One high-complexity AI video: ~4.1 liters water — ~two days drinking water for one person
The Core Insight: Footprints Diverge
Most corporate ESG slides track CO₂ only. UNU finds carbon, water, and land do not move together:
- Switching coal → bioenergy can cut carbon ~70% while raising water footprint 30×+ and land footprint ~100×
- "Green power" for a desert data center may stress aquifers if cooling is evaporative
- Who bears cost matters: Ireland data centers = 21% of metered national electricity (2023); Uruguay saw a drought coincide with a water-intensive DC proposal
That framework explains Kevin O'Leary's Utah retreat the same week — voters rebelled on water, not CO₂.
Our Analysis: Developer and Infra Checklist
1. Site selection = water rights, not just $/MWh
Goldman is pitching $322B SpaceX AI revenue; O'Leary is fighting for 40,000 acres. Both depend on cooling physics. Before you sign a cloud region SLA, ask:
- WUE (water usage effectiveness) disclosure
- Air-cooled vs evaporative mix
- Grid water stress index for the county
2. Model routing saves water, not just tokens
Batch inference, smaller models for drafts, and cache hits cut 945 TWh trajectory the same way they cut Altman's 100B-token bills.
3. Video/gen-AI is the water hog
4.1 L/video means marketing teams generating hero assets have a hidden utility bill. FinOps should add gallons per campaign next to $/1M tokens.
4. Policy tailwind
UNU urges transparency, efficiency, environmental justice across governments, industry, investors, users. Expect EU-style reporting to add water + land columns beside carbon — similar to EU Chips Act sovereignty push but for DC permits.
5. Cross-link power wall
Read AI data center power wall and Big Tech own power plants — water is the other binding constraint.
Key Takeaways
- June 4, 2026: UNU-INWEH report — AI DC water ~9.3T liters by 2030 = 1.3B people's annual domestic water (Sub-Saharan Africa scale)
- Power: 448 TWh (2025) → 945 TWh (2030); ~3% of global electricity; ~400 Mt CO₂eq
- Land: >14,500 km² infrastructure + supply chain footprint
- One AI video ≈ 4.1 L water; one image ≈ 29 mL — gen-media has hidden resource cost
- Low-carbon ≠ low-water — bioenergy and some renewables shift burden to water/land
- For developers: add water-aware routing to FinOps; demand WUE from cloud vendors; treat Utah-style backlash as default for desert builds
- Full report: UNU open access (go.unu.edu/6bRME)
Sources
FAQ
Frequently Asked Questions
How much water will AI data centers use by 2030 according to the UN?
A June 2026 United Nations University (UNU-INWEH) report projects AI-related data center water consumption of 9.3 trillion liters by 2030, equivalent to the basic annual domestic water needs of all 1.3 billion people in Sub-Saharan Africa.
How much electricity will AI data centers consume by 2030?
The UNU-INWEH report estimates global data center electricity use could rise from 448 terawatt-hours in 2025 to 945 TWh by 2030, roughly 3% of world electricity, with associated carbon emissions comparable to the United Kingdom's annual total.
Does low-carbon AI power mean low water use?
No. The UN report warns carbon, water, and land footprints can move in opposite directions — for example, switching from coal to bioenergy may cut carbon about 70% while increasing water use more than thirty-fold and land use about a hundred-fold.
How much water does generating one AI video use?
UNU-INWEH estimates the electricity-associated water footprint of a high-complexity AI-generated video at about 4.1 liters, roughly two days of drinking water for one person, compared with about 29 milliliters for a typical AI image.
Why should developers care about AI water footprint?
Water constraints drive data center siting, permitting, and backlash — as seen in Utah — and gen-AI workloads increase hidden resource costs. Developers should factor water-aware model routing, vendor WUE disclosures, and regional water stress into architecture and FinOps decisions.
Free Weekly Briefing
The AI & Dev Briefing
One honest email a week — what actually matters in AI and software engineering. No noise, no sponsored content. Read by developers across 30+ countries.
No spam. Unsubscribe anytime.
More on Infrastructure
All posts →O'Leary Cuts 40,000-Acre Utah AI Data Center in Half After Backlash
Kevin O'Leary removed 20,000+ acres from the Stratos AI hub in Box Elder County after Utah Senate President Adams and 53% of voters pushed back — water and bird habitat drove the pivot.
IEA Just Released 400 Million Barrels of Oil. It Did Not Work. Here Is What That Means for Tech.
The IEA approved the largest emergency oil release in history after the Strait of Hormuz closed. Brent crude is still above $90. AWS data centers in UAE and Bahrain were hit by drones. Qatar's helium supply is offline, threatening chip fabs globally. Here is the full developer and infrastructure impact.
Ras Laffan Force Majeure: 20% of Global LNG Offline and What It Costs Cloud
QatarEnergy declared force majeure at Ras Laffan — 20% of global LNG offline. Asian LNG hit $25.40/MMBtu. Here's what it means for data center energy costs and AI infrastructure.
Nvidia, Amazon, and Apple Just Closed Their Dubai Offices Because of Iran
Nvidia, Amazon, Apple, and Snap shut Dubai offices as US-Iran tensions ground Gulf flights. Google employees are stranded. Big Tech $50B Middle East AI hub is on pause.
Free Tool
What should your project cost?
Get honest 2026 price ranges for any project type — website, SaaS, MVP, or e-commerce. No fluff.
Try the Website Cost Calculator →Free Tool
Will AI replace your job?
4 questions. Get a personalised developer risk score based on your stack, role, and what you actually build day to day.
Check Your AI Risk Score →Written by
Software Engineer based in Delhi, India. Writes about AI models, semiconductor supply chains, and tech geopolitics — covering the intersection of infrastructure and global events. 816+ posts cited by ChatGPT, Perplexity, and Gemini. Read in 164 countries.