Water, Power, and the Machine: A companion to Truth, Memory, and the Machine
The machine isn’t just code; it’s rivers and grids. If we mean to tell the truth, we have to count the watts and water—and choose tools that keep faith with the world that holds us.
🔧Torchlight Praxis — Tools you can carry.
In Truth, Memory, and the Machine, I argued that “refusing to use AI out of principle may feel righteous, but it risks becoming strategic surrender.” I said the machine could be used for liberation or for erasure—that the ethics were never inside the model, but in our hands.
That was true. It’s still true.
But there’s a cost I didn’t name: the system runs on rivers and grids. It drinks water. It burns electricity. It doesn’t just render images and words; it withdraws from aquifers and leans on power plants that cough into neighborhoods with the least sway to refuse them.
I should have said this sooner. My first piece narrowed the frame to memory, truth, and narrative power. I didn’t trace the pipe to the tap or the wire to the peaker plant. That’s on me. This is the amendment I owe you.
Because this isn’t abstraction. It’s plumbing and wires. It’s heat and drought. It’s the bill a city can’t pay without cutting somewhere else. And if cultural resistance means anything, it has to count those costs before we call our work “freedom.” Culture isn’t just a product of technology. It’s a product of care.
We’ve been trained to chase scale. Faster. Larger. Smarter. In the first essay, I said we can’t afford “purity politics or aesthetic hesitation.” Here, I want to add the other half: we also can’t afford to measure brilliance only by what a tool can do, while ignoring what it takes to do it.
AI at scale doesn’t just mean a big training run somewhere far away. It means everyday usage, everywhere, all the time—millions of queries piling into peak loads and thirsty cooling cycles precisely when a city needs water for people, not servers. The environmental burden isn’t an accident at the edge of innovation; it is the business model of growth.
Let’s look at how this lands.
Memphis, Tennessee. A region sitting on one of the most extraordinary aquifers in the country, wrapped in a history of environmental sacrifice zones. A hyperscale AI build needs power now, not two years from now—so temporary gas turbines spin up to bridge the gap. The air gets heavier. Cooling plans mention wastewater, and maybe that helps, but neighbors still have a right to know: when demand spikes on the hottest day of the year, who runs short first? The answer is rarely the company with lawyers and a statehouse on speed dial.
“...the Health Department needs to do their job to protect the health of ALL Memphians, especially those in frontline communities like 38109,” said LaTricea Adams of Young, Gifted & Green after the permit approval was appealed. (Tennessee Lookout, July 17, 2025).
During the 2023–24 drought in San Antonio, Texas, the pattern repeats. Data centers drew heavily even as neighborhoods tightened their taps; utilities prepared new gas capacity while interconnection requests stacked up; statewide analysts warned that withdrawals tied to data‑center growth could climb sharply in the years ahead; and Hill Country communities weighed gas plants built largely to feed server halls. The press releases spoke of efficiency. The meter spoke of totals.
“While surprise rains generated some green around San Antonio, they did not bring us out of drought. It will take many more steady rainfalls to overcome the last five years of less-than-adequate rain,” the San Antonio Water System wrote on July 1 (San Antonio Water System, July 1, 2025).
Elsewhere, the story rhymes. In Goodyear, Arizona, officials weigh data‑center growth against a finite basin. In Jerome Township, Ohio, permits pause while a township measures cumulative draw. Different climates, same arithmetic: when server halls grow, someone downstream is asked to use less.
That’s the pattern: global promise, local burden. The benefits spread wide. The harms pool where people can least afford to fight them.
And the “solutions” we’re offered rarely touch the root. Efficiency lowers the cost of each computation, which invites more of them. “Water positive” pledges buy projects far from the tap that’s running dry. Air cooling saves water and raises electricity. Liquid cooling reduces electricity and keeps withdrawals high unless fed by non‑potable sources. Siting in cooler, wetter places helps—until incentives and latency targets drag the builds back to hot, dry, politically pliable regions. Transparency is an afterthought. If the numbers were flattering, we’d already have them.
Using AI to run a small, independent press: a fuller accounting
When you run a small press, research doesn’t arrive as a gift. It often comes as a flood—hundreds of pages, half of them noise, the other half scattered across PDFs, hearings, court records, paywalled journals, and posts that may vanish by morning. You learn to listen for signal and build your own nets: library visits, RSS feeds, news alerts, folders, a dozen tabs you promise to close when you’re done. That’s the work before any sentence gets written.
Here’s what the machine can do in that moment. It can sift. It can translate enough of a document to tell you whether it’s worth your time. It can sketch the contours of a topic so you can see the shape of the ground you’re standing on. At its best, used with restraint, it buys you the one thing a small team never has enough of: attention. Not to replace reading, but to make room for it.
But none of that is free. Models are confident when they shouldn’t be. They smooth over the jagged parts of a claim, sand off context, and invent citations that never existed. They reflect the biases they were fed and the power that paid to feed them. And beneath the screen there is the physical cost—the water drawn to keep racks cool, the electricity pulled when the grid is already straining, the neighborhoods that inherit the heat and the hum.
So we keep the older disciplines close. Primary sources first. Reputable secondary analysis next. We ask, every time: Where did this come from? Who benefits if I believe it? Who is missing from this account? We run our fingers along the edge of a statistic until we can feel how it was made—the dates, the method, the exclusions. We verify so that memory doesn’t blur into confidence.
Used this way, the machine is a sorting table, not a printing press. It helps us find and organize, then steps aside. The claims live or die by what the records say, not by what a model summarized about them. We don’t upload anything confidential or identifying. We batch our work, choose the smallest effective tool, and skip the query if we can’t explain why we need it.
There’s a practical rule we hold to, even on deadline. We use these tools when they shorten the distance between readers and the truth—when a translation opens a public document that matters, when a long report becomes navigable enough that we can read the parts that count, when triage lets a small press team keep cadence without sacrificing care. We don’t use them to churn. We don’t use them to replace collaboration that a community can do together. And if the cost to water and power is high and the public interest is thin, we say no.
This isn’t purity or surrender; it’s practice. It is the difference between treating a model as a source and treating it as a map to the sources. It is the choice to keep human judgment—slow, accountable, and named—at the center of how we know what we know.
House Policy
We don’t use AI for drafts or art. We may use it to triage research and open public documents—never as a source—and we verify claims ourselves. If AI helped beyond spellcheck, we’ll add a one-line note at the end.
So, what do we do if the machine is already here and we still need to tell the truth in public?
In the first essay, I wrote: “We don’t win by refusing to fight. We don’t protect the truth by withdrawing from where it’s under attack.” That’s still my position. But here’s the addition: we don’t protect the world by draining it either. Outrage alone won’t save water. Purity alone won’t power a city. We have to practice freedom inside limits we didn’t choose.
And for us, at Torch & Tinder, that means drawing a line.
We have a small staff now—an art director and one staff writer. We still don’t have a full art department or a production team, and the workload is real. But help isn’t free. So here’s the line:
No AI‑generated art. Full stop. Our covers, images, and visual storytelling belong to human hands.
AI‑assisted research is allowed only for discovery and triage: organizing sources, summarizing long records, drafting bibliographies, and translation when it speeds vital access.
Human verification is mandatory. Claims are confirmed against primary or high‑quality secondary sources before publication. AI never stands as a source.
Privacy‑first. No uploading of confidential or personally identifying data.
Minimize footprint. Batch work, prefer smaller models, choose lower‑impact regions/providers when possible, and avoid needless recomputation.
Disclose use with a one‑line note at the end of each piece such as, “AI tools were used to assist with source discovery/translation/summary; all claims were verified and all writing was done and edited by real people.”
We will ask, every time: Why this tool, here, now, at this size? If a smaller model or a slower process will do, we choose that. We batch. We reuse. We avoid needless recomputation. We choose providers and regions that are credibly less harmful when we have the option. If we can’t explain the footprint, we don’t make one.
That’s not perfection. It’s discipline. Presence with restraint. And it’s a correction to our earlier blind spot.
Do the benefits ever justify the draw? Sometimes. When a translation can unlock a safety notice for a neighborhood that’s being lied to. When a dataset becomes a plain‑language guide that keeps a family housed. When a poster or essay gets truth in front of people before the propaganda wave hits. When the work clearly strengthens human hands, not just our sense of cleverness. Sometimes the model is a bridge we can’t build in time any other way.
But “sometimes” is not a blank check. If the use exists to flood a timeline, to chase novelty, or to automate what community can do together—we pass. If we know a region is under drought restrictions or a build depends on temporary gas, we also pass. When we can’t see the costs clearly, we assume they land downstream and use the smallest model lightly—or not at all.
This isn’t anti‑technology. It’s pro‑world.
And it comes with demands.
Demand model‑level reporting on energy and water so cities and communities can make real decisions. Demand siting that favors cooler climates and non‑potable cooling over aquifer draw. Demand community consent and benefit agreements where builds proceed. Demand the smallest effective tool for the task. Demand that we reserve the heaviest models for work that truly cannot be done otherwise. Demand that we count aquifers and air in the price of innovation.
In “Truth, Memory, and the Machine,” I closed with this: “If authoritarians are betting on narrative control through automation—then we bet on presence. On participation. On people who show up, not just to resist, but to shape the future.”
This essay stands beside that one. Our bet is the same—presence, memory, care—but practiced now with the weight of rivers and grids in view. We can use the system. But we choose to use it as if water were scarce and summers were hotter—because they are. We have to build as if the commons is real—because it is. We have to remember, again, that culture is not the product of technology; it is the product of care.
Freedom is a practice. Practice it with memory. Practice it with restraint. Practice it inside the world that keeps us alive.
Explore more from Torch & Tinder Press
📣 Signal Dispatch — Signals from the field
🔧 Torchlight Praxis — Tools you can carry (you’re here)
🔥 Embers — Warmth for the long winter
Stay connected: Instagram · Bluesky · Facebook
Support & community editions: Ko-fi
Please pardon our dust. We’re updating lane names and taglines across the site this week.
Subscribe to Torch & Tinder Press
Books and zines for cultural resistance.
Sources
International Energy Agency. Energy and AI. Paris: IEA, 2025. Executive Summary; “Energy Demand from AI.” https://www.iea.org/reports/energy-and-ai/executive-summary ; https://www.iea.org/reports/energy-and-ai/energy-demand-from-ai
Li, Pengfei, Jianyi Yang, Mohammad A. Islam, and Shaolei Ren. “Making AI Less ‘Thirsty’: Uncovering and Addressing the Secret Water Footprint of AI Models.” arXiv, April 2023. https://arxiv.org/abs/2304.03271
Li, Pengfei, Jianyi Yang, Mohammad A. Islam, and Shaolei Ren. “Making AI Less ‘Thirsty’.” Communications of the ACM, 2025. https://cacm.acm.org/sustainability-and-computing/making-ai-less-thirsty/
Associated Press. “Artificial Intelligence Technology Behind ChatGPT Was Built in Iowa—with a Lot of Water.” September 9, 2023. https://apnews.com/article/chatgpt-gpt4-iowa-ai-water-consumption-microsoft-f551fde98083d17a7e8d904f8be822c4
El País English. “Artificial Intelligence Guzzles Billions of Liters of Water.” November 15, 2023. https://english.elpais.com/technology/2023-11-15/artificial-intelligence-guzzles-billions-of-liters-of-water.html
Walton, Brett. “Data Centers a Small, but Growing Factor in Arizona’s Water Budget.” Circle of Blue, April 2, 2025. https://www.circleofblue.org/2025/supply/data-centers-a-small-but-growing-factor-in-arizonas-water-budget/
“AI Is Taking Water from the Desert.” The Atlantic, March 1, 2024. https://www.theatlantic.com/technology/archive/2024/03/ai-water-climate-microsoft/677602/
“Elon Musk’s xAI Wins Approval for Gas Turbines at Memphis Data Center.” DataCenterDynamics, July 3, 2025. https://www.datacenterdynamics.com/en/news/elon-musk-xai-gas-turbines-memphis/
“Campaigners Appeal Decision to Permit Gas Turbines at xAI’s Memphis Data Center.” DataCenterDynamics, July 17, 2025. https://www.datacenterdynamics.com/en/news/campaigners-appeal-decision-to-permit-gas-turbines-at-xais-memphis-data-center/
“NAACP, Others Appeal xAI Turbine Permits for Memphis Data Center.” Tennessee Lookout, July 17, 2025. https://tennesseelookout.com/briefs/naacp-others-appeal-xai-turbine-permits-for-memphis-data-center/
“Elon Musk’s Memphis AI Data Center Faces Environmental Scrutiny.” TIME, August 2025. https://time.com/7308925/elon-musk-memphis-ai-data-center/
“Ohio Township Pauses Data Center Construction amid Influx of Amazon, Others.” Columbus Dispatch, September 8, 2025. https://www.dispatch.com/story/news/local/2025/09/08/ohio-township-pauses-data-center-construction-amid-influx-of-amazon-others/85989839007/
“San Antonio Data Centers Guzzled 463 Million Gallons of Water as Area Faced Drought.” San Antonio Current, July 30, 2025. https://www.sacurrent.com/news/san-antonio-data-centers-guzzled-463-million-gallons-of-water-as-area-faced-drought-38116670/
San Antonio Water System. “We’re Still in Drought — and Stage 3 Watering Rules.” July 1, 2025. https://www.saws.org/were-still-in-drought-and-stage-3-watering-rules/
“CPS Energy to Acquire Four Gas Plants to Boost Reliability.” Houston Chronicle, September 22, 2025. https://www.chron.com/news/houston-texas/article/antonio-energy-plants-billion-ercot-21060053.php
“CPS Finalizes Purchase of Natural-Gas Plants.” San Antonio Express-News, September 17, 2025. https://www.expressnews.com/business/article/cps-proenergy-acquisition-natural-gas-power-plants-21051049.php
Lewin, Doug. “We’ve Gotta Have It All: Balancing Reliability and Clean Energy on ERCOT.” Texas Energy and Power Newsletter (Substack), 2025. https://www.douglewin.com/p/weve-gotta-have-it-all-balancing
“Texas Is Still in Drought—and AI Data Centers Are Quietly Guzzling up Water.” Austin Chronicle, July 25, 2025. https://www.austinchronicle.com/news/2025-07-25/texas-is-still-in-drought-and-ai-data-centers-are-quietly-guzzling-up-water/
“Texas Data Centers Could Get New Gas Plants to Keep Up with AI.” Texas Tribune, June 5, 2025. https://www.texastribune.org/2025/06/05/texas-data-centers-gas-power-plants-ai/
International Committee of Medical Journal Editors (ICMJE). “Updated Recommendations.” January 2024. https://www.icmje.org/news-and-editorials/updated_recommendations_jan2024.html
International Committee of Medical Journal Editors (ICMJE). ICMJE Recommendations. PDF, 2025 update. https://www.icmje.org/icmje-recommendations.pdf
Associated Press. “Standards Around Generative AI.” August 2023; updated 2024. https://www.ap.org/the-definitive-source/behind-the-news/standards-around-generative-ai/ ; https://www.ap.org/the-definitive-source/behind-the-news/updates-to-generative-ai-standards/
AI tools were used to assist with source discovery; all claims were verified and all writing was done and edited by real people.