During a leisurely school holiday break, I decided to experiment with artificial intelligence by transforming a family photograph into a Norman Rockwell-style painting. Like many others who have dabbled with online AI tools, I was curious to see how technology could reinterpret a personal moment through an artistic lens.
The Environmental Wake-Up Call
My teenage daughter, however, was far from impressed with the digital creation. Beyond declaring the rosy-cheeked results somewhat creepy, she immediately questioned the environmental cost of such AI experimentation. Did I realise how much water was required to power tools like ChatGPT? The truth was, I had never considered it.
This oversight appears to be widespread. American academics studying AI's resource demands suggest the water consumption issue has largely flown under the radar and remained something of an industry secret. Even preliminary investigation reveals concerning statistics about AI's environmental footprint.
Quantifying AI's Thirst
A significant 2023 study from the University of California, Riverside titled Making AI Less "Thirsty": Uncovering and Addressing the Secret Water Footprint of AI Models highlights how AI systems consume water through multiple channels:
- Data centre cooling systems
- Electricity generation for power
- Manufacturing processes for servers
The research projects that global AI demand could account for 4.2 to 6.6 billion cubic meters of water withdrawal by 2027. This staggering volume exceeds the total annual water withdrawal of half the United Kingdom, raising serious concerns amid growing freshwater scarcity worldwide.
The ChatGPT Example
To illustrate the scale, researchers note that ChatGPT needs to drink approximately a 500ml bottle of water for every 10 to 50 medium-length responses it generates. This means using the AI tool to compose just a few emails could consume an entire bottle of water, depending on deployment conditions and location.
Australian Perspectives on AI Sustainability
In Australia, RMIT University senior lecturer Dr Ascelin Gordon addressed the Senate Select Committee into Adopting Artificial Intelligence in 2024, emphasising that environmental impacts of information technology have long been under-recognised.
Dr Gordon revealed that using ChatGPT to generate text consumes between 10 to 90 times more energy per query than a conventional Google search. Furthermore, producing an image through generative AI uses approximately 20 times more energy than generating text, while emerging video generation technology likely requires an order of magnitude more energy again.
These concerns are amplified by the sheer scale of AI usage, with models now running millions or billions of times daily as they become embedded in everyday software platforms.
The Data Centre Reality
The Environmental and Energy Study Institute in the United States reports that larger, AI-focused data centres can consume up to 5 million gallons (nearly 19 million litres) of water daily. This amounts to approximately 1.8 billion litres annually – equivalent to the water usage of a town with 10,000 to 50,000 residents.
Broader Implications of AI Adoption
While AI is frequently promoted for its potential to benefit humanity through efficiency gains, these environmental costs raise important questions about sustainability. Beyond resource consumption, there are concerns about AI's societal impacts, including job displacement and the creation of content that lacks human authenticity.
Noticing AI-generated advertising materials around Canberra, my daughter pointed out their overly perfect yet somehow grim aesthetic. This observation highlights how AI-generated content can sometimes feel soulless compared to human creativity.
A Generational Perspective
As younger generations learn to use AI tools in educational settings, developing critical thinking about their environmental and social impacts becomes increasingly important. My daughter's awareness of AI's significant shortcomings represents exactly the kind of thoughtful engagement needed as these technologies become more pervasive.
While my Norman Rockwell experiment was entertaining, understanding its hidden water cost has changed my perspective. The environmental implications of casual AI use deserve greater public attention and responsible industry action to address this growing sustainability challenge.
