The Case for Stewardship

The problems are measured. Our answers are intentions — held to the same standard.

Every figure sourced and dated · Last checked October 2026

Seven problems, side by side with what we intend to do about them.

The problem, measured

Every number on the left comes from a government agency, a national laboratory, a grid monitor, a peer-reviewed study, a national pollster, an independent tracker, or the company that reported it. Each one links to its source at the bottom of the page.

Our intended step, designed

Everything on the right is design intent, not a result. Node Zero isn’t built yet, and nothing here claims an outcome the data hasn’t shown.

How we’ll know

Each step names what we’ll measure and publish. Measure first, claim second. If the evidence doesn’t support a step, we don’t take it.

01

Computing’s appetite for power is outrunning the grid.

The problem · measured
4.4%

of all U.S. electricity went to data centers in 2023 — 176 terawatt-hours. By 2028 that is projected to reach 6.7% to 12%.1

$6.3B

of the $16.4 billion in capacity charges from PJM’s 2028/2029 capacity auction, held in July 2026 — 38.2% — came from data center load, according to PJM’s independent market monitor. PJM is the grid operator for much of the eastern U.S.2

$14–$37

a month: how much a typical Dominion Energy household in Virginia could see its generation and transmission costs rise by 2040 because of data center demand, in 2024 dollars.3

U.S. data center electricity use: about 60 terawatt-hours in 2014, 76 in 2018, 176 in 2023, and a projected 325 to 580 in 2028.~6020147620181762023325–5802028 (projected)Terawatt-hours per year, U.S. data centers. 2028 is a projected range.
U.S. data center electricity use, terawatt-hours.1
Our intended step · designed, not yet proven

Grow only on evidence.

Node Zero’s computing is modular and staged. Each module is meant to be added only after the last one has produced reviewable performance data — not built out to a forecast and paid for by the neighbors. The energy strategy stays subject to utility review and agreement.

How we’ll know: energy use and performance reported publicly, module by module, before any expansion.

02

Data centers are drinking from the same water we do.

The problem · measured
66B

liters of water consumed on site by U.S. data centers in 2023, up from 21.2 billion in 2014 — plus nearly 800 billion more used to generate their electricity.1

776.5M

gallons consumed in 2024 by Google’s data center site in Berkeley County, South Carolina — about 2.1 million gallons a day.4

U.S. data center water: 21.2 billion liters consumed on site in 2014, 66 billion in 2023, and nearly 800 billion more used indirectly through electricity generation in 2023.On-site water consumed, 201421.2 billion LOn-site water consumed, 202366 billion LUsed indirectly to make its electricity, 2023~800 billion L
U.S. data center water use, billions of liters.1
Our intended step · designed, not yet proven

Use water that’s already there, and protect it.

Sealed compute is designed to sit in an already-flooded quarry pit so that keeping it cool doesn’t draw on the water a town drinks. That is the benefit we are designing for: drinking water left alone. We don’t claim energy savings over the best-run land data centers; the published record doesn’t show them. Do no harm comes first: the pit is studied before any hardware enters it.

How we’ll know: water used per unit of computing reported — including what evaporates from the pit — and temperature, dissolved oxygen, pH, turbidity, and conductivity measured before expansion and published. If the evidence doesn’t support the next phase, there is no next phase.

03

Communities are saying no — and they have reasons.

The problem · measured
71%

of U.S. adults oppose an AI data center being built in their area — 48% strongly. Fewer, 53%, oppose a nuclear power plant in their area.5

~$130B

in data center projects — at least 75 of them — blocked or delayed by local opposition in the first quarter of 2026 alone, by one tracker’s count.6

Gallup, March 2026: 71 percent of U.S. adults oppose an AI data center in their area, compared with 53 percent who oppose a nuclear power plant in their area.Oppose an AI data center in their area71%Oppose a nuclear power plant in their area53%
Share of U.S. adults opposed, March 2026.5
Our intended step · designed, not yet proven

Built with the community, or not built.

The Stewardship Engine starts by opening the doors and convening the community’s own institutions — schools, co-ops, civic and tribal leaders, workers, and families — as participants, not an audience. A site moves forward only with the support of the community around it. Without that, the mission dies, and it should.

How we’ll know: a site moves forward only with visible community support, decided in the open — and the campus stays open for anyone to walk through and read the data.

04

The jobs are few. The tax breaks are large.

The problem · measured
~50

full-time workers at a typical 250,000-square-foot data center once it’s running — about half of them contractors — after roughly 1,500 at the peak of construction.3

$928M

in tax savings the industry received from Virginia’s data center sales and use tax exemption in a single year, fiscal 2023.3

Virginia JLARC: about 1,500 workers on site at the peak of construction, and about 50 full-time workers once a typical data center is running.Workers on site at peak construction~1,500Full-time workers once it runs~50
Workers at a typical Virginia data center, as reported to JLARC by industry representatives.3
Our intended step · designed, not yet proven

Benefit that stays where it’s made.

The campus is designed to need people: skilled jobs building, running, researching, and caring for the site, paid a wage a person can build a life on. And it’s designed to be shared — rental eco cabins, community gardens and farmers markets, trails, demonstrations, and supervised water-based recreation, open to the public. Schools and universities get a living campus as a classroom.

How we’ll know: who works on the site, what they’re paid, and what stays in the region, reported alongside the water data.

05

Almost every watt becomes heat — and almost none of it is used.

The problem · measured

Virtually all the electricity a data center uses leaves it as heat. Where it is finally being measured, almost none of that heat is put to use.

1.9%

of the heat generated by the EU data centres that reported in the first round of mandatory reporting (filed 2024, covering 2023) was reused — and the European Commission cautions the data is incomplete. Only 67 reporting data centres reported any heat reuse.7

We could find no national figure for the United States at all.

European Commission, first reporting period: about 1.9 percent of the heat generated by reporting EU data centres was reused.1.9%of the heat reused
Share of heat reused by reporting EU data centres, first reporting period.7
Our intended step · designed, not yet proven

Treat heat as part of the place.

The campus is planned as one living system, so the heat its computers produce is part of the design from the beginning: accounted for and kept within do-no-harm limits for the water. We make no claim that the heat will be reused; in South Carolina’s warm climate, most of it arrives when nobody needs it.

How we’ll know: a year-round baseline and a heat budget for the pit before any load, then water temperature, oxygen, and stratification watched continuously against that baseline, with limits set before operation starts.

06

Extraction leaves holes, and the land and water around them pay for it.

The problem · measured
22,500+

known abandoned hardrock mine features on federal land alone — pits, tunnels, and more — pose or may pose environmental hazards, out of at least 140,000 known features.8

~500

permitted mine sites in South Carolina today — sand, clay, granite, limestone, kaolin, and more.9

Dozens

of people are injured or killed every year across the country exploring or playing on mine property.10

Our intended step · designed, not yet proven

Put the damaged places to work, and make them safe to visit.

Stewardship Compute is purpose-built for flooded quarries and abandoned mine pits — already-disturbed water, with no new excavation. Around the pit, the campus is designed to give the public a cared-for way to enjoy the land and the water: trails and docks, with swimming and water recreation only where trained lifeguards are on duty and the water is tested. Unsupervised quarry swimming is dangerous, and we won’t pretend otherwise.

How we’ll know: every Node sited only on already-disturbed land, with its water and land condition measured before work begins and reported after — plus recreational water quality, lifeguard hours, and every incident logged and published.

07

Our rivers and lakes are already struggling.

The problem · measured
47%

of U.S. river and stream miles were in poor biological condition, judged by their bottom-dwelling insect life, in EPA’s 2018–19 national survey.11

50%

of U.S. lakes were in poor condition for phosphorus in 2022 — a nutrient that can feed algae blooms.12

U.S. EPA surveys: 47 percent of river and stream miles in poor biological condition (2018 to 2019), and 50 percent of lakes in poor condition for phosphorus (2022).River and stream miles in poor biological condition47%Lakes in poor condition for phosphorus50%
U.S. EPA National Aquatic Resource Surveys.1112
Our intended step · designed, not yet proven

Measure the water, in public, before claiming anything.

Do no harm first — then, maybe, do good. Whether a stewarded site can go beyond doing no harm and actually leave its water healthier is the open question our field research exists to answer. We won’t claim it before the data does.

How we’ll know: a baseline study before hardware enters the water, continuous monitoring after, and data anyone can read.

The numbers we won’t use.

Some of the most-repeated figures about data centers can’t be traced to a primary source when you go looking for where they came from. We left them out.

  • “A large data center uses up to 5 million gallons of water a day.” Widely repeated; we couldn’t trace it to a primary measurement. We used a real, reported site instead.
  • “Every ChatGPT conversation drinks a bottle of water.” The study behind it estimated about 500 milliliters for roughly 10 to 50 medium-length responses from GPT-3, depending on where and when it runs — an estimate its authors call conservative.13

Where every number comes from.

  1. Lawrence Berkeley National Laboratory (Shehabi et al.), 2024 United States Data Center Energy Usage Report, December 2024. DOI 10.71468/P1WC7Q. escholarship.org (PDF)
  2. Monitoring Analytics, Independent Market Monitor for PJM, Market Monitor Report, Members Committee webinar, July 27, 2026. www.pjm.com
  3. Virginia Joint Legislative Audit and Review Commission, Data Centers in Virginia (Report 598), December 9, 2024. jlarc.virginia.gov
  4. Google, 2025 Environmental Report, “Alphabet’s Schedule of Water Use by Data Center Location” (Berkeley County, SC). The per-day figure is our own arithmetic. www.gstatic.com
  5. Gallup, Americans Oppose AI Data Centers in Their Area, May 13, 2026 (survey of U.S. adults, March 2–18, 2026). news.gallup.com
  6. Data Center Watch, Q1 2026 Data Center Watch Report. A private tracker’s estimate; “blocked or delayed” combines both outcomes and is not broken out. www.datacenterwatch.org
  7. European Commission, DG Energy, Assessment of the energy performance and sustainability of data centres in the EU — First technical report, July 2025. op.europa.eu
  8. U.S. Government Accountability Office, Abandoned Hardrock Mines (GAO-23-105408), January 2023. www.gao.gov
  9. South Carolina Department of Environmental Services, Mining and Reclamation program. www.des.sc.gov
  10. U.S. Mine Safety and Health Administration, MNM Safety Alert: Dangers Exist at Active and Abandoned Mine Sites! (Stay Out–Stay Alive campaign), June 2016. www.msha.gov
  11. U.S. EPA, National Rivers and Streams Assessment 2018–19: Key Findings. www.epa.gov
  12. U.S. EPA, National Lakes Assessment 2022: Key Findings. www.epa.gov
  13. Li, Yang, Islam & Ren, “Making AI Less ‘Thirsty’,” Communications of the ACM 68(7), 2025. arxiv.org

Figures checked against their sources in October 2026. If you find an error, tell us through the contact page and we’ll correct it in the open.

The problems are already measured.

Our answers aren’t proven yet. Node Zero is where we find out — in the open.