The Hidden Water Cost of Artificial Intelligence
PHOENIX – May 24,2024 –
The expansion of artificial intelligence is causing a meaningful surge in fresh water consumption,due to the cooling needs of data centers. From training large language models to processing user requests, AI’s thirst for water is growing rapidly, posing environmental and economic concerns. This has led to a need for more enduring practices.As the technology continues to evolve, further analysis will be needed to address its impact.
The Hidden Water Cost of Artificial Intelligence
Artificial intelligence is rapidly transforming industries, but its expansion carries a meaningful environmental result: a growing demand for fresh water. Each interaction with AI models like ChatGPT consumes water, and these seemingly small amounts add up quickly.
Key Takeaways
- Training a single large-language model such as ChatGPT can consume hundreds of thousands of liters of fresh water.
- Data-centre electricity demand is expected to surge 16% by 2030, amplifying water-cooling needs.
Water: AI’s Silent thirst
AI chips generate substantial heat, necessitating cooling systems in data centers. Most large-scale facilities use evaporative cooling towers, which consume clean water and release it as steam. Researchers estimate that training ChatGPT alone vaporizes about 185,000 gallons of water. A typical user session, involving 10 to 50 prompts, uses about half a liter of water.
Goldman Sachs forecasts a 165% increase in data-center power capacity by 2030. This surge will intensify the cycle of energy demands, heat generation, and water consumption.
Why It’s an Environmental Concern
Fresh, clean water is a precious resource, and approximately one-fifth of data centers are located in water-stressed regions. These centers compete with drinking water supplies and agriculture. In Phoenix, Arizona, data centers can demand over 170 million gallons of water daily for cooling, exacerbating regional water shortages.
Heavy water use depletes aquifers, while discharging warmer effluent can alter river temperatures and harm ecosystems. Climate change further complicates the issue, with hotter summers increasing cooling demands as droughts reduce water reserves.
Fast Fact
Could pig poop ponds hold the answer to AI data center water usage? Companies specializing in filtering contaminants, including pig sewage, are proposing the repurposing of waste or low-quality water to reduce reliance on fresh groundwater.
How AI’s water Use Stacks Up
Global AI demand is projected to consume 1.1 trillion to 1.7 trillion gallons of freshwater annually by 2027. This rivals the annual household water use of California and is growing faster than any sector outside agriculture.
Semiconductor fabrication plants, known for their high water consumption, can use up to 10 million gallons daily, equivalent to the needs of a midsize U.S. city. Hyperscale data centers are rapidly catching up, with some now using over 5 million gallons daily, rivaling towns of 50,000 residents.
Agriculture accounts for about 70% of global groundwater use. However, in drought-prone, high-income regions, AI’s water consumption directly competes with farms, households, and manufacturers, increasing the likelihood of usage caps, taxes, or charges.
Tip
In addition to water, electricity demands from the AI sector may more than double this decade, potentially forcing utilities to restart shuttered plants or import pricier renewables, increasing costs for consumers.
What Can Be Done Before the Well Runs Dry?
Water-intensive AI firms face scrutiny from regulators and environmentally conscious shareholders. Venture capital is flowing into projects for efficient immersion cooling,membrane recycling,and leak-detection platforms for data centers. Investors can explore established cooling-tower manufacturers or water-themed ETFs like Invesco’s Water Resources ETF (PHO) or First Trust’s Water ETF (FIW).
Due diligence when considering AI companies should include specific metrics such as water-use efficiency, the hydrological risk of data-center locations, and progress toward “water-positive” pledges, alongside customary AI growth metrics.
The Bottom Line
The race to dominate generative AI is increasingly linked to a growing water bill. If unmanaged, the conflict between AI and water could reduce margins, invite regulatory and stakeholder opposition, reshape site-selection considerations, and harm fragile water ecosystems worldwide.
Investors who consider the hydrological balance sheet-and support companies that curb, recycle, and monetize every drop-will be better positioned when “liquidity scarcity” shifts from warnings to reality.
