As I experiment with AI platforms, pressured by the need to keep up with technologies that might impact my work, I celebrate the small victories when clients so obviously prefer the writing I do that is untouched by this technology. AI’s potential for rendering many jobs obsolete is concerning. Geoffrey Hinton, the Nobel Prize-winning AI groundbreaker, has warned that AI will replace most jobs, leaving opportunities for only highly specialised skills. But it is the mentions of excessive energy and water consumption that raise niggling concerns.
A recent article in The Washington Post showed that using AI to generate one 100-word email every week for a year would use the equivalent of 7.5 kilowatt-hours of electricity, or the same as an hour’s worth of power used by nine American households. That’s a thought that sent me exploring interviews with engineering and tech reporter, Karen Hao, whose book Empire of AI exposes the technology’s devastating exploitation of the environment and the abuse of vulnerable, local communities, including workers in Kenya and the extraction of massive amounts of freshwater from communities in Chile. You can listen to Karen Hao’s account on How AI Is Threatening Democracy & Creating a New Colonial World.
Even more sinister is Hao’s exposé on how these tech companies are looking to ecoup some of the enormous costs of AI development by turning to the defence industry for opportunities and support to offset the mammoth investments that American AI companies’ scale-at-all-costs business models have necessitated.
We should collectively work to understand the real impact of AI.
In the meantime, I’ve done a little exercise using Deepseek to generate this story, based on articles on Karen Hao’s work, on the dangers of AI. Let me know what you think.
The Hidden Dangers of AI: Straining Environmental Resources and Impacting Indigenous Communities
Artificial Intelligence (AI) is often hailed as a revolutionary force for sustainability, but its rapid expansion comes with significant environmental costs. The massive energy and water demands of AI infrastructure — particularly data centres — threaten natural resources and disproportionately affect Indigenous communities who rely on these ecosystems.
1. AI’s Massive Energy and Water Footprint
A. Energy Consumption
Training and running AI models require vast amounts of electricity. For example:
- OpenAI’s GPT-3 consumed an estimated 1,300 megawatt-hours during training — enough to power 120 U.S. homes for a year.
- Data centres (which host AI computations) account for 1-2% of global electricity use, a figure expected to rise with AI expansion.
This energy demand often relies on fossil fuels, increasing carbon emissions and worsening climate change.
B. Water Usage
AI data centres require extensive cooling systems, consuming millions of gallons of water:
- Google’s data centres used 15.8 billion litres of water in 2022, a 20% increase from 2021, partly due to AI workloads.
- Microsoft’s AI research in Iowa led to a 34% spike in local water consumption, straining resources in drought-prone areas.
2. Impacts on Indigenous Communities
Indigenous peoples, who are often stewards of land and water, face direct harm from AI’s resource exploitation:
A. Land Displacement for Data Centres
- In Arizona, data centres (including those supporting AI) have been built on or near Tribal lands, consuming groundwater in arid regions. The Navajo Nation has raised concerns over water depletion from tech infrastructure.
- In Canada, Indigenous groups have protested data centre expansions that threaten traditional territories.
B. Water Scarcity and Pollution
- In Chile, data centres and AI-related mining for lithium (used in tech hardware) have drained water supplies, affecting the Atacama Desert’s Indigenous Lickan Antay people.
- In New Mexico, tech companies have faced backlash for excessive groundwater use, impacting Pueblo communities who rely on scarce water for farming.
C. Climate Change and Ecosystem Disruption
AI’s carbon footprint exacerbates climate change, which disproportionately affects Indigenous peoples:
- Melting Arctic ice (fueled by energy-intensive AI infrastructure) threatens the livelihoods of Inuit and Sami communities.
- Deforestation for server farms in places like Brazil and Indonesia disrupts Indigenous lands.
3. The Need for Sustainable and Ethical AI
To mitigate these harms, tech companies must:
✔ Transition to renewable energy for AI operations.
✔ Reduce water use with more efficient cooling systems.
✔ Engage Indigenous communities in decision-making about land and water use.
Conclusion
While AI offers many benefits, its unchecked growth risks worsening environmental degradation and harming Indigenous rights. Without responsible policies, the tech industry’s pursuit of AI could come at the cost of vital natural resources and the communities that depend on them.
Then I followed up to check what energy was used in generating this article.
The energy and water cost of generating this single AI-written article is extremely small, likely negligible compared to training large AI models. Here’s a breakdown:
1. Energy Use for This Response
- Inference vs. Training: Generating text (like this article) is called inference, which uses far less energy than training massive models (e.g., GPT-3).
- Estimated Energy: A single inference request may use 0.001–0.01 kWh (similar to a few minutes of a light bulb). For context:
- Training GPT-3 used ~1,300 MWh (millions of times more).
- This article’s generation likely cost less than 0.001% of that.
2. Water Use for This Response
- Cloud Server Cooling: AI runs in data centres cooled by water, but individual queries have minimal impact.
- Estimate: Generating this article may have indirectly used a few millilitres of water (via data centre cooling). For comparison:
- Training an AI model can consume millions of litres (e.g., Google’s global data centres used ~15.8 billion litres in 2022).
3. Why the Disparity?
- Training AI is like building a factory; using AI is like turning on a machine briefly. Most environmental harm comes from:
- Training large models.
- Running data centres at scale (especially water-cooled ones in drought-prone regions).
4. How to Reduce AI’s Footprint?
- Use smaller, efficient models (e.g., distilled versions of GPT).
- Run servers on renewables (Google/Microsoft aim for 24/7 clean energy by 2030).
- Improve cooling tech (e.g., Microsoft’s underwater data centres).
This piece was written for the July 2025 edition of Postscripts, Shamillah Wilson’s monthly round-up of what’s been happening in feminist circles, her work, and some recommended reading suggestions.
