The Data Center Problem Is a Human Problem, and It Needs a Human Answer

Let’s start with Cofiwch Dryweryn!

In 1955, the City of Liverpool decided it needed more water.

Sixty miles away, across a national border, there was a valley in North Wales called Tryweryn, and in it a village called Capel Celyn. Forty-eight people, a dozen farms, a chapel, a school, a post office, a graveyard. One of the last places in Britain where an entire community spoke Welsh and only Welsh.

Liverpool did not apply for planning permission in Wales. If it had, every Welsh authority with standing could have objected, and every one of them would have. Instead, Liverpool Corporation took the scheme directly to Parliament in Westminster as a Private Bill — an entirely legal route, and that meant no Welsh planning body ever got a vote.

When the bill came up, thirty-five of the thirty-six Welsh MPs voted against it. The thirty-sixth abstained. Not one voted in favor. A hundred and twenty-five Welsh local authorities opposed it.

It passed.

The locals hired a bus, drove to Liverpool, and marched through the city center carrying signs. They were ignored. Over the next nine years the farms were bought out, the chapel was demolished, and the bodies in the graveyard were exhumed and moved. The reservoir opened in 1965. The opening ceremony was abandoned partway through because of the protest.

In dry summers the water drops and the valley comes back — the stone walls, the bridge, the line of the lane. People drive out to look.

In 2005, forty years later, Liverpool City Council formally apologized.

The valley is still under water.

On a ruined wall near Llanrhystud, someone painted three words in the 1960s. They have been repainted by hundreds of hands ever since. Cofiwch Dryweryn. Remember Tryweryn.

Everything about that story was legal. Every procedure was followed, the compensation was paid, and Liverpool really needed the water. Nobody in it is a cartoon villain.

And almost everyone now agrees something went badly wrong — not the reservoir, but the way the decision got made.

Hold onto that. We are about to do it again, at continental scale.

The Building You Have Not Noticed

Chances are there is a building near you that has slipped past your attention. Windowless and plain, behind a fence, a handful of cars in the lot, a substation humming quietly beside it.

That building is why your phone works. It is why your show streams without a hitch, why your doctor can pull scans from a distant hospital in seconds, why your bank balance reconciles, why your car finds the fast route. It is where researchers in Ohio solve protein-folding problems that used to take decades.

Our whole civilization leans on these buildings, and almost nobody thinks about them.

That is ending, because the consequences have arrived.

The Scale Nobody Prepared For

Here are the numbers, and where they come from — because this is a field where wrong numbers travel fast.

Global data centers consumed roughly 415 terawatt-hours of electricity in 2024: about 1.5% of all the electricity on earth. The International Energy Agency projects that to roughly double by 2030, to around 945 TWh, or just under 3% of global supply. That is approximately what the entire country of Japan uses in a year. Demand is growing about 15% annually — more than four times faster than every other sector of electricity demand combined.

The growth rate is the story, not the current total.

Water is harder to see. A single large evaporatively cooled facility can consume one to five million gallons a day, the household draw of a town of ten to fifty thousand people, and roughly 85% of it evaporates rather than returning to the local supply. In Northern Virginia, the densest cluster on earth, reported use approached two billion gallons in 2023; Loudoun County alone represented for about 899 million.

But here is the figure almost nobody publishes: the water consumed generating the electricity is often as large as, or larger than, the water used on site. A facility on a closed-loop cooling system can report near-zero water and still carry an enormous footprint upstream at the power plant. When an operator gives you a water number, ask which number.

On prices: U.S. residential electricity rates rose about 25% between 2020 and 2024, while rates for large commercial users stayed roughly flat. That gap is real. Whether data centers caused it is honestly disputed. Several studies find no clear link between data center density and current retail prices, and a large share of that increase traces to the natural gas price spike of 2021–22, not to server racks. Meanwhile, wholesale and capacity-market prices — which move much faster than regulated retail rates — have risen sharply in data-center-dense regions.

Reasonable people looking at the same data are reaching different conclusions. That is not a reason to stop paying attention. It is a reason to demand better instruments.

And the argument is already producing policy. In late 2025, Virginia regulators approved a rate structure assigning a larger and firmer share of service costs to data centers specifically, so that other ratepayers carry less of it.

What the Numbers Leave Out

Four things that rarely make the coverage, and that change the moral arithmetic.

The tax bill is often forgiven. Virginia’s data center tax abatement now runs roughly $1.6 billion a year. Georgia revised its FY2026 abatement cost projection upward by 664%, to about $2.5 billion. Between 2016 and 2025, four large tech owners in Oregon took $616 million in property tax abatements. A community can host the infrastructure and still not collect from it.

The permanent jobs are few. Construction hires thousands. Operations do not. Georgia Tech researchers found that rural facilities typically employ fewer than one hundred permanent workers and import specialized services from outside the community. Both halves of that are true, and both need saying.

The land, the air, and the noise are real. A hyperscale campus is measured in hundreds of acres. Diesel backup generators come with air permits. Low-frequency mechanical noise is the most common complaint from neighbors — the kind of harm that is easy to dismiss and impossible to live beside. And servers turn over every three to five years, which makes electronic waste a structural output rather than an accident.

Someone is doing the invisible work. Behind every model are data annotators and content moderators, largely in Kenya, India, and the Philippines, earning roughly $1.46 to $3.74 an hour for work that pays $21 to $27 in the United States, with thoroughly documented psychological injury from prolonged exposure to violent material. Kenya’s draft 2026 AI policy now mandates psychosocial support, written contracts, and transparent pay reporting. That is a labor problem — but it is not a labor shortage.

And the strongest case for the defense, which deserves stating plainly: compute displaces other resource consumption: fewer commutes, tighter logistics, faster drug discovery, better-optimized grids, remote diagnosis. Efficiency per unit of computation has improved by orders of magnitude. The honest position is not that these buildings are bad. It is that we are not counting properly.

Why This Revolution Is Different

We have done this twice.

The agricultural revolution began roughly twelve thousand years ago with domestication and unfolded during millennia. The industrial revolution reorganized human life around the engine over about a century and a half. Both were severe. Both were also slow enough that institutions, laws, schools, and guilds might bend around them. Grandparents and grandchildren lived different lives, but there were generations in between to absorb the shock.

This one — call it the autonomous revolution, the shift to systems that decide and act without a person in the loop — is arriving in years.

The problem may not be that the change is bigger. It is that our institutions were built for a slower world. Permitting a power plant takes a decade. Training an electrician takes four years. A treaty takes longer than both. A company can announce a campus and break ground before the county has finished reading the filing.

The crisis is not the buildings, and it is not the chips. It is the widening gap between how fast one thing moves and how fast the other can.

And a gap that size cannot be closed locally. Water, zoning, and electricity are local. Compute, capital, supply chains, and the atmosphere are not. One country regulating alone does not solve the problem. It relocates it somewhere with fewer rules.

That is why the answer has to be global, however unfashionable that is to say in 2026.

Innovate

Engineering is currently ahead of politics, and there is real room left.

Modern cooling can cut cooling energy by up to half. Liquid immersion reduces cost by about 40% and water use by up to 90% against older methods. An MIT spin-out, borrowing from nuclear engineering, gained another 15% in power efficiency through mastering how bubbles form and release from server surfaces: practical physics, practical impact.

Waste heat is not a problem. It is an unclaimed asset. In Odense, Denmark, a campus delivers captured server heat to roughly 11,000 homes; a Microsoft site at Høje-Taastrup heats about 6,000. The EU now requires large facilities to assess heat recovery. Ireland is building its first district heating system around an AWS facility. A building that warms a town is a different kind of neighbor than one that only draws.

Points worth pushing on:

• Treat heat reuse as a siting criterion, not a retrofit. Co-designing is cheap. Retrofitting is not.
• Match new load with new generation — and say which generation. Behind-the-meter gas, restarted nuclear, small modular nuclear reactors, and long-duration storage are all on the table, and they are not equivalent.
• Fund the unglamorous chokepoints: transformers, high-voltage gear, interconnection queues. They never make headlines, and they decide everything.
• Site for climate, not for tax breaks. A drought basin and a water-rich cold region are not the same country, whatever the abatement schedule says.
• Push efficiency at the model layer. The greenest kilowatt-hour is the one never required.

Educate

Here is the part almost nobody outside the industry knows: we do not have enough people.

Industry estimates put the construction shortfall as high as 499,000 workers, with calls for more than 300,000 new electricians. Electrical work alone is 45–70% of construction cost. Nearly 30% of union electricians are between 50 and 70, with roughly 20,000 retiring a year — 200,000 out the door over a decade against a pipeline filling a fraction of it. More than 340,000 data center positions are currently open. These figures come from different sources with different definitions and horizons; treat them as a direction, not a decimal.

Pause on what they mean. The most advanced technology in human history is bottlenecked on people who know how to pull cable and terminate a bus duct.

This is the most hopeful fact in this essay. These are not jobs lost to AI. They are jobs created by it, in exactly the trades we spent forty years steering young people away from.

But it has to be said in the same breath as the annotators. A shortage of electricians in Ohio and a surplus of underpaid moderators in Nairobi are the same economy making two different decisions about whose labor is worth protecting.

Points worth pushing on:

• Rebuild the trades pipeline at the speed of a national mission: apprenticeships, community college partnerships, employer-funded training with real wage floors. The companies putting up the campuses should fund the workforce that makes them possible.
• Stop equating “AI education” with learning to code. The industry needs electricians, HVAC techs, controls specialists, commissioning engineers, high-voltage crews, water systems engineers, and grid operators — and it needs ethicists and philosophers.
• Extend labor standards to the annotation and moderation supply chain. Kenya has started. It should not have to finish alone.
• Teach civic literacy about infrastructure. Every commissioner voting on a rezoning should know what a load factor is, what water usage effectiveness measures, and what a power purchase agreement actually obligates. Most do not. That is fixable, and cheap.
• Grow the pipeline everywhere. Africa holds about 18% of the world’s people and roughly 5.5% of its data centers — and its share of installed capacity, measured in megawatts, is smaller still, because a hyperscale campus and a single colocation room both count as “one.” Only about 5% of Africa’s AI talent can reach the compute it needs. That is not a market gap. It is structural exclusion.

Collaborate

So who should lead? The honest answer is: different actors on different things.

The EU has regulatory experience and is already setting rules others will copy. The Nordics have proven heat reuse at scale. The US and China have the capital and the build velocity. India brings demand growth and standing in the Global South. Japan and South Korea hold the hardware. Africa has the fastest-growing need and the least infrastructure, and no arrangement that leaves Africa out will hold.

Nobody has to lead on everything. Coalitions of the willing and able beat a grand bargain nobody signs.

The pieces are assembling. The European Commission has a Data Center Energy Efficiency Package alongside its digital and AI roadmap. The industry’s Climate Neutral Data Center Pact targets carbon-neutral facilities by 2030. The Pact is a voluntary commitment, and the binding obligations are narrower. There are serious proposals to bring AI compute into the Paris Agreement framework, discussed at India’s AI Impact Summit, with mandatory emissions reporting and coordination through the UNFCCC, OECD, and G20. The UN’s Global Digital Compact calls for North–South cooperation on affordable compute, though the mechanisms remain undeveloped.

Points worth pushing on:

• Measure before you regulate. A shared, audited reporting standard for energy, water, emissions, and grid impact at each facility comes first. You cannot negotiate over numbers nobody agrees on. And be clear about what today’s metrics do not tell you: PUE says how efficiently a building delivers power to a chip. It says nothing about whether the computation was worth doing.
• Prevent the race to the bottom. If strict jurisdictions push facilities toward weak ones, the global impact does not fall. It moves, usually somewhere with less recourse. Floors matter more than ceilings.
• Build shared compute access: regional facilities, public research capacity, compute for countries that cannot build their own. The alternative is an environment in which most people are data sources rather than participants.
• Anchor it in institutions that already exist — UNFCCC, OECD, G20, IEA, IEC. New ones take a decade. There isn’t a decade.
• Give communities real standing, not a comment period after the fact. More than a hundred American communities have already enacted moratoriums, and over three hundred state data-center bills were filed in the first six weeks of 2026. People are not waiting to be given a voice. They are taking one.

What This Actually Requires

Every problem here — the grid, the water, the workforce, the patchwork of rules, the gap between nations with compute and nations without — is a problem made by people. None of it is a law of nature. All of it is the residue of choices, most of them made in rooms, by people who believed they were being responsible.

Machines will not negotiate a water-sharing agreement. They will not fund an apprenticeship in a town that lost its factory. They will not decide that a village in Kenya should have access to compute. Those are decisions, and decisions require someone willing to look past the next quarter and the next election.

We do not have millennia this time, or even a century and a half. We have a handful of years to build institutions that can keep pace with something already moving faster than they are.

If you take three things from this:

Find out where your nearest data center is, who approved it, and whether there was a public hearing. Ask any operator you can reach what their water usage effectiveness is — and whether that figure includes the water used to generate their electricity. And when the next campus is announced near you, show up before the vote rather than after it. And make sure you know what is truly the weakest link in the situation.

The AI Ship has sailed. There is no turning back. Only going forward, but hopefully with know-how, guardrails, policies, measurements, and behaviors to produce the greater good for all humans.

Innovate. Educate. Collaborate.

Technology has always been the easy part.

In 2005, Liverpool apologized. It was a real apology, and it cost the council something to make.

The valley is still underwater.

Cofiwch Dryweryn!

---

Note:

I have spent my career in emerging technology with some of the most amazing companies. I’m an AI and Emerging Technology Strategist and techno-public philosopher who has explored how technology forms our society and culture. AI has been at the center of my conversations and experience, working on the ground floor of emerging technologies, including machine learning, artificial intelligence, and internet technology. Something a friend taught me and drilled into my mind was this: “Every situation, no matter how complex it initially looks, is exceedingly simple.” ~ Eli Goldratt

---

Sources
I have spent a good part of the summer researching this for a class I am teaching this fall. This blog gives you a glimpse of my Fall class with the added insights of World Religions, Thomas Merton, and Catholic Social Teaching.

Energy and water

• International Energy Agency — Energy and AI (2025); Electricity 2026.
• EESI — Data Centers and Water Consumption.
• Environmental Law Institute — Data Centers and Water Fact Sheet, January 2026.
• Center for Secure Water, University of Illinois — Data Center Expansion in Virginia: Closing Essential Gaps for Knowledgeable Water Planning and Permitting.
• Bay Journal — As data centers multiply in the Chesapeake region, water use increases too.

Electricity prices and rate design

• E3 — Understanding the Drivers of Rising Electricity Rates and the Role of Data Centers (2026).
• Independent Institute — Have Data Centers Actually Raised Electricity Prices?
• Yale Climate Connections — Home electricity bills are skyrocketing.
• Inside Climate News — Virginia Regulators Approve New Dominion Rates, Assign More Costs to Data Centers, January 2026.

Fiscal impact and local government

• Good Jobs First — Even Cloudier with a Greater Loss of Spending Control: How Data Center Tax Abatements Weaken Public Budgets.
• Cardinal News — Tax abatement for data centers is now $1.6 billion a year.
• Brookings — New evidence on data center employment effects.
• Georgia Institute of Technology research on rural data center employment.

Workforce

• Build.inc — Data Center Construction Labor Shortage 2026.
• Introl — 340,000 Unfilled Data Center Jobs.
• Institute for Human Rights and Business — Content moderation is a new factory floor of exploitation.
• Business Daily Africa — AI giants face new minimum pay, mental healthcare rule in Kenya.
• CHI 2026 — “The plan is just survival”: Data Work in Kenya and the Regime of Entrapment.

Engineering and heat reuse

• MIT News — Nuclear-inspired cooling system.
• Microsoft Local — Surplus datacenter heat in Denmark; Munters — Odense district heating case study.
• CNBC — Data centers, AI and district heating in Ireland.

Governance

• White & Case — EU data center energy regulatory outlook 2026; Climate Neutral Data Center Pact.
• NUS Center for International Law — An international agreement on AI climate sustainability.
• CSIS — From Divide to Delivery: How AI Can Serve the Global South.
• UNOSSC — From AI Divide to AI Dividend, HLPF 2026.

Tryweryn

Nation.Cymru — Cofiwch Dryweryn: A nation remembers Tryweryn.

National Library of Wales — The Drowning of Tryweryn Valley.

BBC News — Tryweryn a “shameful chapter in Welsh history”, says minister, October 2015.

---

Eli Goldratt PhD

The Goal: A Process of Ongoing Improvement (if you can get the Third Edition with case studies, get it)

Critical Chain (if you need help managing projects)

Necessary But Not Sufficient (one of my favorites)

The Choice (This one is NOT a book about TOC, but one of the best by Eli)

Eliyahu M. Goldratt was an Israeli physicist turned management philosopher, educator, and business consultant. Although trained in physics, he became one of the most influential thinkers in operations management during the late 20th century. His work fundamentally changed how many organizations think about productivity, improvement, and systems.  His work has been widely applied across industries, from manufacturing to project management, to education and healthcare, and even banking. The work has influenced corporate strategies worldwide. Goldratt was known as a provocative thinker who challenged conventional business, cultural, and social assumptions, often described as a “slayer of sacred cows” 

He is best known for creating the Theory of Constraints (TOC) and for writing the influential business novel The Goal. He founded the Avraham Y. Goldratt Institute and developed educational programs, including the Goldratt Satellite Program and TOC Insights, to teach TOC principles interactively (Theory of Constraints Institute tocinstitute.org). He passed away in 2011 from complications of lung cancer, leaving a lasting impact on management theory and practice