The Microdose

The Environmental Impact of LLMs

AI promised unlimited growth. But it's hitting real-world constraints: water, power, and land.
Adam Wildheart
collage for environmental impact of AI LLMs: girl wearing a daisy flower crown with OpenAI, Anthropic, and Deep Seek logos

© The Microdose

Subscribe to The Microdose
Subscribe -- post
Cheri Wildheart
Adam Wildheart

AI promises infinite growth. But it’s hitting real-world constraints: water, power, and land.

These are constraints you can’t code away. We need to innovate past them, or face significant environmental costs.

AI's real-world limits

AI models are power hungry. It takes a lot of water and energy to run chatbots and LLMs.

  • Every AI-generated email uses about 1 bottle of water for cooling.

  • Most data centers don’t disclose water use, and two-thirds are built in water stressed areas. [Washington Post]

  • Data centers drive demand for gas-powered electricity, increasing pollution and fossil fuel dependency. [MRT]

  • Big tech is heavily investing in solar, but it may not meet future demand. [TechCrunch]

These aren’t theoretical problems. They’re the real impact of “growth at all costs.”

Emerging solutions

  • Advanced cooling tech (like targeted thermoelectric chips) can drastically cut water and energy needs. [MIT]

  • Next-gen optical computer chips use light instead of electricity, significantly reducing heat and cooling requirements.

  • Massive solar projects are rapidly expanding to offset AI data-center energy demands. [TechCrunch]

  • Advanced nuclear power could offer clean, scalable energy – but it may take a decade or more to safely roll out. [Bloomberg]

  • Space-based AI. China is experimenting with orbit-based AI data centers. Satellites radiate heat directly into space, need no water, are solar-powered, and take up zero land. No constraints.

Good news! AI is pivoting to efficiency

  • Smaller, specialized AI models require far less data, energy, and water.

  • Next-gen chips and cooling systems slash power and water use dramatically.

  • Efficiency is profitable. Sustainable AI isn’t just good for Earth; it’s good for business. [MIT]

AI isn’t moving backward. It’s pivoting to smarter, cleaner growth.

Here's how to stay ahead of AI's resource crunch

  • Build for efficiency from day one. Use smaller LLM models where you can. Efficiency saves resources and costs immediately.

  • Turn sustainability into your selling point. Efficiency makes your product attractive, profitable, and future-proof.

  • Plan for a space-friendly future. AI will soon leave today’s constraints behind. Think ahead about innovation, including space-based opportunities.

AI data centers face real-world constraints, but that doesn’t have to limit your ability to innovate.