DEPLOYDatabase

AI data centers

Are AI data centers bad for the environment?

Power + siting on the record for tracked campuses

The impact is real, but it is not one number. It comes down to two things we track: how a data center gets its power, and where it is built.

A data center's carbon comes from the electricity it uses. So a site on the local grid is only as clean as that grid, a site on its own solar or nuclear power is much cleaner, and a site on its own gas engines burns new fossil fuel.

The other big factor is the land it sits on.

The big country-wide numbers on this page come from outside research.

What we track is how each site we follow gets its power and where it is built, which is what really decides the answer.

Last verified not yet on record2 sources

Find out instantly when it changes.

Verified changes only. We check the record and link every source. No marketing.

There is no one answer, and any page that gives you one is not being honest. The truth is that a data center's impact comes down to two things, and we track both.

The first is power.

Computers turn electricity into work and heat, so a site's carbon is really the carbon of the electricity it uses.

A site on the local grid is exactly as clean or dirty as that grid.

A site on its own solar and batteries, or a nuclear deal, is much cleaner.

A site running its own gas engines is burning new fossil fuel instead of using the grid, and that is the biggest worry. The table below sorts every site we track by which of these it uses.

The second is land and water.

A site built on open land takes ground that was never built on. A site built on an old factory or industrial lot reuses land that was already used up, which is easier on the area.

Water is covered on its own page. Some sites use a lot, some almost none. We record the land and the water source for each site.

One thing we do not have is an exact carbon number for a single site.

The big numbers, like US data centers using about 4.4 percent of the country's electricity in 2023 and maybe 6.7 to 12 percent by 2028, come from a national lab, not from us. This page checks the things that decide the impact, and points to the research for the totals.

How the sites we track are powered and built

Across 103 AI data centers we track, this is what really decides the impact: the power source and the land, site by site, plus 121 power and generation projects we follow.

Low-carbon powered
7
On the grid
59
Behind-the-meter gas
17
On reused industrial land
12

What is claimed, and who says it

Each box says who the fact comes from. The last box is the honest limit of what we can answer.

The main impact (from LBNL)

Electricity use, growing fast

US data centers used about 4.4 percent of the country's electricity in 2023, and that could reach 6.7 to 12 percent by 2028, mostly because of AI.

That demand, and the carbon behind it, is the heart of the worry.

The number is from a national lab. We track the sites behind it.

Why there is no single grade

Carbon follows the power

A data center makes almost nothing at the building itself. Its carbon is really the carbon of the electricity it uses.

So the same building is dirty on a coal grid and clean on solar or nuclear.

That is why we sort the sites we track by their power, instead of giving one grade for the whole industry.

What we have, and what we do not

The causes, not the exact carbon

We track how each site is powered and where it is built, the two things that decide the impact, shown below.

We do not have an exact carbon number for a single site, and we do not claim one.

The big totals come from outside research.

Clean power: solar, storage, or nuclear

Runs mostly on its own solar and batteries, or a deal to buy nuclear power. This is the cleanest kind on our record.

7 tracked campuses

Grid power

Pulls power from the local grid, so it is as clean or dirty as that grid is.

59 tracked campuses

59 tracked campuses
AWS Madison County Data Center CampusesCantonMISOprivate
AirTrunk JHB1 Johor Bahru Data CentreJohor Bahruotherprivate
Amazon 'Bridgefield' GWU Ashburn CampusAshburnPJMbrownfield
Amazon (AWS) Hermiston / Umatilla Data Center CampusHermistonWECCgreenfield
Amazon Project Rainier Data Center (New Carlisle)New CarlislePJMprivate
Barber LakeColorado CityERCOTprivate
Black PearlWinkERCOTprivate
ByteDance Pecem Data Center Campus (TikTok)Pecemother-

On-site gas

Runs on its own gas engines, which burn new fossil fuel instead of using the grid. This is the biggest carbon worry on our record.

Mixed or other

A mix of sources, or a setup that does not fit one clean group.

20 tracked campuses

20 tracked campuses
Alterra Marshall Technology & Energy CenterMarshall-greenfield
Amazon (AWS) Boardman Data Center CampusBoardmanWECCgreenfield
Data City (Energy Abundance)LaredoERCOTgreenfield
Fermi America HyperGrid (Project Matador)AmarilloSPPprivate
GW Ranch (Pacifico Energy)Fort StocktonERCOTprivate
Google / DTE 'Project Cannoli' — Van Buren TownshipVan Buren TownshipMISOgreenfield
Google Visakhapatnam AI Hub (Adani-Google Data Centre Campus)Visakhapatnamotherprivate
Matrix Data Center Campus (The Matrix Center)Sulphur SpringsERCOTgreenfield
How we know this

The big electricity numbers here come from a US national lab, and the way water use works from a federal report. The power, grid, and land facts for each site come straight from that site's record on DEPLOY.

We check how each site is powered and where it is built, which is what decides the impact, and we point to the research for the big totals. We do not have an exact carbon number for a single site, and we do not claim one here.

Common questions

Do AI data centers produce a lot of carbon?
Only as much as their electricity does. A data center makes almost no carbon at the building itself. Its carbon is really the carbon of the power it uses. So a site on a coal-heavy grid has a big footprint, one on solar or nuclear a small one, and one on its own gas engines burns new fuel. We record how each site we track gets its power, which is the thing that decides this.
How much electricity do data centers use?
US data centers used about 4.4 percent of the country's electricity in 2023, and that could reach 6.7 to 12 percent by 2028, according to Lawrence Berkeley National Laboratory. Most of that growth is from AI. That is the heart of the environmental worry. We track the individual sites behind it and how each one is powered.
Are data centers destroying farmland?
Land is the real question here, and we track it. A site on open land takes ground that was never built on, which is what the farmland worry is about. A site on an old factory or industrial lot reuses land that was already used. We sort the sites we track by their land type, so you can see it instead of guessing.
Are AI data centers using renewable energy?
Some are. We mark sites that run on their own solar and batteries or a nuclear deal as clean power, and we track the power and generation projects tied to sites. We also mark the ones on their own gas engines. The page shows the split instead of making one claim about the whole industry.
What is the biggest environmental concern with data centers?
The three most common worries are electricity use and its carbon, cooling water, and land and local effects. All three change from site to site, which is why we track power, water, and land for each one instead of giving a single grade. Water has its own page.

Keep reading

Ask the registry

Track this answer

Find out instantly when it changes.

Verified changes only. We check the record and link every source. No marketing.

Canonical URL: /are-ai-data-centers-bad-for-the-environment · machine-readable mirror: /are-ai-data-centers-bad-for-the-environment.md