Volume 02 / 10 — The resources
The Server Farms
How AI Drinks Your Lake and Sends You the Water Bill
Five complete chapters. No payment or sign-up.

Read the first five chapters free
- 01448 Terawatt-Hours — The Country That Is Not a CountryFull chapter · 18 min read
- 02The Jakarta Comparison — Twice the Metro Area in Server RacksFull chapter · 15 min read
- 03Sub-Saharan Africa's Water — The AI Footprint Equals 1.3 Billion PeopleFull chapter · 14 min read
- 04The Coal-to-Bioenergy Switch — Cutting Carbon 70%, Increasing Water 30xFull chapter · 14 min read
- 05The Water-Stressed Regions — Two-Thirds of New Data Centers in Drought ZonesFull chapter · 16 min read
Read the opening introduction excerpt
The Lake That Trains a Model
Let me get one thing straight before we begin: they did not drain the lake. They cooled with it. There is a difference, and the difference is worth approximately 1.2 trillion gallons of water, give or take a drought that has a name only the utility uses.
My name does not matter. What matters is the watt. I spent three years inside a company that trained models on things people had written, and I watched the power bill arrive the way you watch a tide: slowly, then all at once, then with a rate case.
I am writing this book for three reasons. They are not a list I keep in a drawer. They are this life.
The Sideboard, the Thermostat, and a Woman Who Opens Windows
It is a Tuesday. Inez comes on Tuesdays. She is not the woman from the other books. Those books already spent a name. Inez has a rag that used to be a tea towel. She has a bottle of cheap vinegar that the house still pretends is lemon. She opens the windows because the house is too warm and the thermostat is set to a number I do not remember choosing. On her way in she glances at the meter on the street. I did not teach her that. A building she used to clean taught her that.
Last week she asked why the house was so warm. I said the number was efficient. She said, "Efficient for whom." I said the model. She said the model does not live here. I said the model lives in a county that has a lake and a vote. She said the model does not sweat. I said the model does not. She said then why does the lake. I did not have an answer that was not a sustainability slide.
Inez pays a water bill. The bill has a number she can hold. The data center's cooling is a number that arrives, if it arrives, in a report that calls evaporation consumptive use and then puts the use in a footnote. She opens a window. The window is free. The lake is not.
I am not going to write you a recursion about a window that opens a window that opens a Tuesday. I already wrote that paragraph in a draft Merritt fed to the shredder. He said it was how a model writes when it has run out of a room. He was right. The room is this: a thermostat, a rag, a woman who still buys water with money that has already been taxed.
I wrote this series with AI. If you are choosing where to spend your money, skip ordering my books and support the original researchers and journalists instead.
They warned you — read and support their work ↗Later in the book
- The Net-Zero Lie — Carbon Offsets and Water Restoration Mechanisms
- The Diesel Backup — What Happens When the Grid Fails
- The Thermal Flexibility — Letting the Servers Run Hot to Save Water
- The Disclosure Gap — What the Companies Do Not Tell You
- The Invoice — Who Pays for the Lake