THE DISPATCH · WORLD 05 / 05 · UNBROKEN PLANET
THE OPEN KNOWLEDGE PROJECT
Unbroken Planet, the Series, and the Bill the Machines Are Running Up
Every answer a machine gives you was paid for somewhere. In megawatt-hours. In gallons. In a line on somebody's power bill in Ohio. This dispatch puts the current public numbers on the table with their sources attached, then explains the open library being built so that record does not end up owned by the same people selling the machines.
FIGURES CITED FROM PUBLIC SOURCES · UNBROKEN PLANET IS LABELED FICTION · THE OPEN LIBRARY AND COMMUNITY ROOMS ARE IN BUILD, NOT OPEN
MEDIA · OFFICIAL UNBROKEN PLANET ARTWORK · SUPPLIED BY BONNER BOLTON · NOT A DOCUMENTARY PHOTOGRAPH
I GOT ASKED WHAT THE HORSE COST.
Nobody asks what the model cost. Not the subscription — the cost. The ground it sits on, the river it drinks, the transformer it waits in line for. That question is the whole reason Unbroken Planet exists.
I have spent most of my life around animals and equipment that tell you immediately when you are asking too much of them. A horse goes sour. A truck throws a rod. The feedback is fast, physical, and impossible to spin. Software has no such tell. You type, it answers, the cost lands three states away on somebody who never typed anything.
So this is not an anti-AI dispatch. I build with these tools every day, and Arrowhead Seven is built on them. It is a receipts dispatch. Below are the figures that are actually public as of September 2026, each one attached to where it came from, followed by the honest other half of the argument, and then the thing I am asking you to help build.
FEATURED · ROBERT EDWARD GRANT · THIRD-PARTY WORK, CREDITED
THE UNIVERSAL RELATIONSHIPS OF √2, φ, π & e.
DIAGRAM · © ROBERT EDWARD GRANT · SHARED WITH ATTRIBUTION · NOT AN UNBROKEN PLANET OR BONNER.AI WORK
Why it is here: the Open Knowledge Project keeps receipts on what machines cost, and Grant’s argument is about what the record itself is made of — a claim that the constants we build on are one structure, not several. Agree or not, it is the kind of first-principles question this dispatch exists to hold in public. The conversation below goes deeper.
FEATURED CONVERSATION · YOUTUBE (EMILIO ORTIZ, 2026-08-22, 1:54:45) · THIRD-PARTY VIDEO, NOT AN UNBROKEN PLANET PRODUCTION · LOADS ONLY WHEN YOU PRESS PLAY
THE FIRST NUMBER
THE GRID IS THE REAL CONSTRAINT.
The International Energy Agency put the first credible global number on this. Data centres consumed about 415 terawatt-hours of electricity in 2024 — roughly 1.5% of the world's electricity — and the IEA projects that roughly doubles to about 945 TWh by 2030, just under 3% of global supply. The line that matters is not the total. It is the slope: about 15% per year, more than four times faster than every other sector of electricity demand.
READ IT HONESTLY · The same IEA analysis notes data centre growth accounts for less than 10% of global electricity demand growth to 2030. The strain is not evenly spread — it is concentrated, local, and lands hardest on specific grids. That is a different problem than a planetary one, and it deserves the more accurate description.
THE SECOND NUMBER
IT DRINKS. AND IT DRINKS WHERE PEOPLE LIVE.
Cooling is the quiet half of the story, and it is the half I take personally, because in Texas water is not an abstraction. It is whether the tank fills. The Environmental and Energy Study Institute's compilation of federal and utility figures puts U.S. data centre direct water use at roughly 449 million gallons a day across some 5,426 facilities — and that is before the water burned to generate the electricity, which runs far larger.
A single large facility can pull up to 5 million gallons a day. Whether that is catastrophic or trivial depends entirely on the watershed it sits in, which is exactly why a national average is the wrong unit and a county-level record is the right one. I could not find one public, usable county-by-county record that joins those numbers. That is a gap you can actually close.
THE THIRD NUMBER
SOMEBODY IS ALREADY PAYING FOR IT.
This is the part that stopped being theoretical. PJM is the grid operator for thirteen states and D.C. Its capacity auction — the market that pays generators to guarantee they can show up on the worst day of the year — cleared at $28.92 per megawatt-day for 2024/2025. For 2025/2026 it cleared at $269.92. For 2026/2027, $329.17.
In Ohio, IEEFA cites an estimate that capacity-market costs add about sixteen dollars a month to an average residential bill; its market-monitor source attributes much of the auction-price increase to projected data-centre demand. That is a system cost, not a per-prompt invoice — and it should not be invisible.
THE HONEST OTHER HALF
THE CASE AGAINST MY OWN ALARM.
If I only printed the frightening numbers I would be doing the same thing I am accusing the industry of — selling a story instead of showing a ledger. So here is the strongest version of the other side, from the operators' own published reporting.
EFFICIENCY
THE MACHINES GOT BETTER FAST
Google reports a fleet-wide PUE overhead of 0.09 against an industry average of 0.54 — roughly 83% less overhead energy per unit of compute. Efficiency gains have repeatedly outrun demand growth before.
OFFSET
EMISSIONS AND DEMAND CAN DIVERGE
Google's 2026 report states data centre electricity demand rose 37% in 2025 while operational emissions fell 2%, against nearly 35 GW of clean-energy agreements signed since 2010. Supply-chain emissions still rose 25%.
UPSIDE
THE TOOL CUTS SOME OF ITS OWN BILL
Google attributes 41 million tCO2e of avoided emissions in 2025 to AI-enabled solutions across nine products. Whether avoided-emissions accounting is rigorous is a fair fight — but it is on the record and it is auditable.
MY POSITION · Both halves are true at once, and the industry's own disclosures are the best evidence for both. What is missing is not data. It is an independent, comparable, public place to put it where the people affected can read it without a subscription and without taking a vendor's word for the framing.
THE ANSWER I CAN ACTUALLY BUILD
THE OPEN KNOWLEDGE PROJECT.
The Open Knowledge Project is being built as a public, source-attributed record of what building with AI actually costs and actually produces — kept by the people doing the building, in the open, in a form a county commissioner or a rancher or a sixteen-year-old can read.
Not a think tank. Not a manifesto. A ledger. Every entry answers the same five questions, and an entry that cannot answer them does not go in.
01 / 05
WHAT WAS MADE
The artifact, the brief behind it, and the human judgment that changed the output. Result first, method second.
02 / 05
WHAT IT COST
Tools used, passes run, measured spend, and energy where it can be honestly estimated. Unknowns stated as unknowns, not rounded to zero.
03 / 05
WHERE IT LANDED
The ground it touched. Which grid, which watershed, which community. Local, not national averages.
04 / 05
WHO OWNS IT
Provenance and license recorded at creation. A fingerprint and a timestamp support a record — they do not by themselves prove authorship or ownership, and this project will not claim otherwise.
05 / 05
WHAT IT GOT WRONG
The failed passes and the retractions stay in the record. A ledger you can only add wins to is marketing.
STATE OF THE PROJECT · Honest label, because that is the whole point: the Open Knowledge Project is in build, not open. There is no public submission form, contributor agreement, moderation rail, or published corpus today. Founding members are being gathered now so the terms, attribution rules, and safety review get written with the first contributors instead of handed to them. Anything on this page describing the library in the present tense is describing intent, not an active service.
WHY IT COMES WRAPPED IN A STORY
UNBROKEN PLANET IS THE SERIES.
Unbroken Planet is a fictional story world built by Arrowhead Seven — a western set in a wild near future where the land is still open and the machines have learned to speak. It is clearly labeled fiction. It is not documentary footage and it is not a key to my real life, which lives under my own name in the rest of The Dispatch.
Here is why the two are bolted together. A spreadsheet about capacity auctions changes no one's mind. A story about what a place is worth might. The series is the front door; the Open Knowledge Project is the workshop behind it. Every episode is produced with the same AI-assisted tools this dispatch is scrutinizing — which means the production itself is the first entry in the ledger. If I am going to ask an industry to show its receipts, I go first.
THE WORLD
UNBROKENPLANET.AI
The story world's own home. Releases and characters live there, labeled as fiction, on their own schedule.
THE WORKSHOP
ARROWHEAD SEVEN
The studio and Creative Intelligence Lab where the series is actually made — and where each production's costs get logged.
THE STANCE
KEEP THE EARTH WILD
Use the smallest tool that does the job. Return more than you take. The full framework is in High Tech, Wild Places.
LATEST FROM THE FIELD
WHAT IS MOVING RIGHT NOW.
Current reporting in these areas, linked to the source so you can check my reading against theirs. Headlines are the publishers' own.
- Energy demand from AI — Energy and AIInternational Energy Agency · the base global dataset
- Demand — Electricity 2026International Energy Agency · data centres as ~50% of U.S. demand growth to 2030
- Projected data center growth spurs PJM capacity prices by factor of 10IEEFA · who pays, and how much
- Data Centers and Water ConsumptionEnvironmental and Energy Study Institute · direct and indirect water
- Read Google's 2026 Environmental ReportGoogle · operator disclosure, both halves of the argument
- Data center moratoriums are not a substitute for oversightBrookings · the policy fight now reaching county level
- The AI Boom Is Stressing the Grid—but It Doesn't Have to Be This WayNRDC · the mitigation case
- AI: Five charts that put data-centre energy use — and emissions — into contextCarbon Brief · the anti-hype corrective, worth reading against my own framing
METHOD · Figures on this page were read from the linked sources on September 2, 2026 and are quoted as those publishers state them. Where estimates are contested — per-prompt water use especially — the estimate is labeled as an estimate. If you find an error here, that correction is itself a contribution, and it will be credited.
AUDIENCE SIGNAL · FIVE QUESTIONS
WHERE DO YOU STAND IN THIS?
Not a personality test and not a diagnosis. It is a routing question. Your answers tell me which direction of this work deserves the next field report, and they show you which lane of the Open Knowledge Project you would actually be useful in. Answers are aggregated. Nothing here builds an advertising profile.
YOUR LANE
RESULT
Only your resulting lane — not your individual answers — is sent as an aggregate analytics signal to steer what gets reported next. That event is not attached to the Field Letter waitlist or a member identity. Results are not medical, financial, or professional advice, and no answer here is treated as a commitment. Privacy & data.
THE ASK
BE A FOUNDING CONTRIBUTOR.
The Open Knowledge Project only works if the record is kept by more than one person, and the terms get written with the first contributors rather than handed to them. The quiz lane is an aggregate signal only; it is not attached to the Field Letter waitlist or a member account. The contribution rail is not open. If it opens, attribution, licensing, moderation, and contributor terms will be published before submissions are accepted.
THIS DISPATCH EARNS
OPEN-KNOWLEDGE STARTER KIT
The five-question entry template above as a working checklist, plus the source list from this page in a citable format. Access opens with the approved member sign-in rail; no public submission or community-review path is active yet.
SOURCES
CHECK MY WORK.
- International Energy Agency — Energy demand from AI, Energy and AI. 415 TWh (2024), ~1.5% of global electricity, ~945 TWh projected 2030, 15%/yr growth, U.S. +240 TWh.
- International Energy Agency — Demand, Electricity 2026. Data centre expansion as roughly 50% of U.S. demand growth to 2030.
- Institute for Energy Economics and Financial Analysis — Projected data center growth spurs PJM capacity prices by factor of 10. $28.92 to $269.92 to $329.17 per MW-day; 63% attribution; $9.3B consumer cost.
- Environmental and Energy Study Institute — Data Centers and Water Consumption. 163.7B gal/yr direct; ~211B gal indirect (2023); Northern Virginia +63%; UC Riverside per-prompt estimate.
- Google — 2026 Environmental Report. Data centre demand +37% (2025), operational emissions -2%, supply-chain emissions +25%, PUE overhead 0.09, 41 MtCO2e AI-enabled avoided emissions.
- Carbon Brief — Five charts that put data-centre energy use into context.
- Brookings — Data center moratoriums are not a substitute for oversight.
- NRDC — The AI Boom Is Stressing the Grid—but It Doesn't Have to Be This Way.
NO AFFILIATION · None of these organisations sponsor, endorse, or are affiliated with Bonner.AI, Arrowhead Seven, or Unbroken Planet. They are cited as sources only.
WEEKLY FIELD LETTER · WAITLIST REQUEST
Series launch early notice is included. Confirm once, keep the EPISODES topic on, and the Unbroken Planet launch note reaches this list first.
Request a place on the future Field Letter waitlist for Dispatch stories, field maps, release notes, and audience-voted topics. This newsletter request does not create member access, record your quiz lane, or enroll you as a contributor. Email consent stays pending until confirmation can be delivered and completed; delivery stays held until the approved mail rail is active.
EMAIL CONFIRMED · DELIVERY HELD.
YOUR EMAIL is confirmed, but no Field Letter is delivered until the approved mail rail is active.