AI Data Centers: Why Everyone's Fighting About Them, and What Makes One Clean

Data centers used to be invisible. They were anonymous gray buildings off a highway exit, and nobody outside the industry thought about them.

AI changed that. Data centers are now front-page news, local election issues and the subject of packed town hall meetings. People want to know what’s being built down the road, how much power and water it will use, and who pays for it.

Here’s a plain-English guide: why AI data centers are such a big deal right now, what they actually use, and the five things that separate a clean data center from a dirty one.

The short answer

• AI is driving a data center boom. US data centers went from about 4.4% of the country’s electricity in 2023 to a projected 6.7–12% by 2028.
• Communities are pushing back. Higher power bills, water use and tax breaks are the top complaints. Billions of dollars of projects were blocked or delayed in 2026.
• Not all data centers are equal. Where a data center gets its power and how it cools itself matter more than almost anything else.
• A clean data center runs on a renewable grid, barely uses water, wastes little energy, reuses its heat and is open about all of it.

Why data centers are suddenly everywhere

Every AI answer is computed in a data center, on specialized chips that use a lot of power and run hot. As AI went from novelty to everyday tool, companies started building data centers faster than ever.

The numbers are big:

• United States: data centers used about 176 TWh in 2023, around 4.4% of all US electricity, according to a Department of Energy report. By 2028 that could reach 6.7% to 12%.
• Worldwide: the International Energy Agency estimates data centers used about 485 TWh in 2025 and expects roughly 950 TWh by 2030, about double.

Per question, AI is small. Google measured a median Gemini text prompt at 0.24 watt-hours and 0.26 mL of water. The problem is scale: billions of prompts a day, plus training ever-larger models.

Why towns are saying no

A single large AI campus can use as much power as a small city. That lands on real places, and people there have real concerns:

• Power bills. If a utility builds new plants and lines for a data center, households worry they’ll share the cost.
• Water. Many data centers evaporate water to cool their chips, sometimes in places already short of it.
• The grid. New demand can delay coal plant retirements or bring in new gas plants.
• Noise, land and tax breaks. Big buildings, constant fan hum, and incentives that can outweigh the jobs created.

The result is a wave of local opposition. Data Center Watch counted at least 75 US projects, worth about $130 billion, blocked or delayed in the first quarter of 2026 alone, the most it had ever recorded in a single quarter.

Most of that opposition isn’t anti-technology. It’s about who carries the cost. And that comes down to how a data center is built and powered.

What a data center actually uses

Electricity goes to two places: the computers themselves, and everything that keeps them running, mainly cooling.

The industry measures the overhead with PUE (power usage effectiveness): total electricity divided by the electricity used by the computers. A PUE of 1.0 would mean zero overhead. The Uptime Institute’s 2025 survey put the global average at 1.54, a number that hasn’t moved much in years. So the typical data center spends about a third of its electricity on overhead.

Water is mostly for cooling. The cheapest way to cool chips in a hot climate is to evaporate water, the same way sweat cools you. That water is gone. Some designs recirculate water in a closed loop instead, or use cold outside air, and use far less.

The five things that make a data center clean

1. A grid that actually runs on clean power

This is the big one. The same server running the same AI model can have a tiny or a large carbon footprint depending on the grid it plugs into. A data center on a coal-heavy grid and one on a hydro and geothermal grid are worlds apart.

There are two ways a company can call its AI “renewable”:

• Clean by location: the data center sits on a grid that runs on renewables, so the electricity it draws is clean.
• Clean by matching: the data center uses whatever the local grid provides, and the company buys or funds renewable energy elsewhere to cover the same amount.

Matching is better than nothing, especially when it funds new renewable projects. But clean by location is the gold standard, because the power going into the chips is clean.

2. Cooling that doesn’t drink the local water supply

In a cold climate, outside air can do most of the cooling for free. Closed-loop liquid cooling recirculates the same water instead of evaporating it. Either way, a clean data center uses a small fraction of the water of an evaporative one in a hot, dry region. Read more in how much water does AI use?

3. Low overhead (a low PUE)

The less energy spent on cooling and power conversion, the more goes into useful work. Cold-climate data centers can run close to 1.0, far below the 1.54 global average.

4. Waste heat put to use

Every watt a chip uses becomes heat. Most data centers vent it. The best pipe it somewhere useful: district heating, greenhouses, fish farms.

5. Openness

A clean data center isn’t a secret. You should be able to find out where your AI runs, what powers it and how it’s cooled. If a company won’t say, assume the answer isn’t flattering.

Where Viro AI runs

We built Viro around these rules. Viro AI, Viro’s own model, runs in data centers powered by 100% renewable energy, clean by location, not by offsets:

Location	Power	What stands out
Akureyri, Iceland	Hydro and geothermal	Liquid cooling and cold Atlantic air; waste heat warms a community greenhouse
Blönduós, Iceland	Hydro and geothermal	The Arctic climate does most of the cooling: PUE 1.03
Keflavík, Iceland	Hydro and geothermal	On the same clean Icelandic grid
Glomfjord, Norway	Hydropower	Beside hydro dams on the Arctic Circle; fjord water cools the racks in a closed loop, and waste heat is headed for nearby fish farms
Dublin, Ireland	100% renewable electricity	Viro’s EU home, so European chats are processed in the EU
Helsinki, Finland	Hydro and wind	Renewable power on a cold-climate site

Across these sites, PUE runs from 1.03 to 1.31, so nearly all of the electricity goes into computing. You can see them on a map at Viro’s data centers.

What about ChatGPT, Claude and Gemini? Viro offers them too, and they run in their makers’ own data centers, which we don’t control. Viro labels them Energy matched and funds new renewable energy to cover the electricity your chats with them use. We say that plainly because it’s a real difference.

The cleanest data center is the one you don’t need

There’s a third option: no data center at all. Small AI models can now run on your own phone or laptop. In the Viro Mac app, you can choose Apple Intelligence and the answer is computed on your Mac, on your own power, with no cooling water. We call it bring your own data center.

What you can do

1. Use the smallest model that does the job. Rewrites and quick questions don’t need a giant model.
2. Run small tasks on your device when you can.
3. Choose AI that tells you where it runs, and prefer clean by location over matching.
4. Ask your local officials what new data centers will plug into and how they’ll be cooled.

AI isn’t going away, and neither are data centers. But the difference between a clean one and a dirty one is huge, and it’s a choice.

For more, read is AI bad for the environment? and what makes AI eco-friendly, or try Viro AI free.

Frequently asked questions

How much electricity do data centers use?

In the US, data centers used about 4.4% of all electricity in 2023, according to a Department of Energy report, and could use 6.7% to 12% by 2028. Worldwide, the International Energy Agency estimates about 485 TWh in 2025, heading toward roughly 950 TWh by 2030.

Why are people against AI data centers?

The main worries are higher electricity bills, water use for cooling, strain on the local grid, noise, land use and generous tax breaks. In the first quarter of 2026 alone, Data Center Watch counted at least 75 US projects worth about $130 billion blocked or delayed by local opposition.

What makes a data center clean?

Five things: it plugs into a grid that actually runs on renewable power, it cools without evaporating lots of water, it wastes little energy on overhead (a low PUE), it puts its waste heat to use, and it publishes where it is and how it's powered.

What is PUE?

Power usage effectiveness: a data center's total electricity divided by the electricity its computers use. 1.0 would mean zero overhead. The global average has sat around 1.54 for years, according to the Uptime Institute; cold-climate sites can get close to 1.0.

Where are Viro's data centers?

Viro AI, Viro's own model, runs in Akureyri, Blönduós and Keflavík in Iceland, Glomfjord in Norway, Dublin in Ireland and Helsinki in Finland, on grids powered by hydro, geothermal and wind.

Do ChatGPT, Claude and Gemini run in Viro's data centers?

No. They run in their makers' own data centers. Viro labels them Energy matched and funds new renewable energy to cover the electricity your chats with them use.

Sources

1. DOE Releases New Report Evaluating Increase in Electricity Demand from Data Centers, US Department of Energy, December 2024
2. Key Questions on Energy and AI, International Energy Agency, 2026
3. Q1 2026 Data Center Watch Report, Data Center Watch, 2026
4. Uptime Institute Global Data Center Survey 2025, Uptime Institute, July 2025
5. How much energy does Google's AI use? We did the math, Google Cloud, August 2025
6. Our Data Centers, Viro Help Center

Cover photo by Jeremy Bishop on Unsplash