AI Data Centers in 2026: Power, Water and UK Growth
AI data centers are specialized buildings packed with powerful chips that train and run AI models. They look like normal data centers from outside. Inside, they are very different, and they need far more electricity and cooling.
Here is what makes them special, what they cost the planet, and why Britain is suddenly building so many.
What is an AI data center?
A regular data center stores files, runs websites, and handles emails. An AI data center does something heavier. It runs thousands of GPUs (graphics chips) that work together on one giant task.
Think of a normal data center as a big library. An AI data center is more like a power station with a library attached. Every part is built to feed those chips with power, data, and cold air.
How are they different from normal data centers?
The gap is huge. A Hugging Face explainer by Boris Gamazaychikov and Sasha Luccioni points out three big ones:
- Chip power: A new Nvidia Blackwell GPU can draw up to 600 watts. A recent regular CPU draws about 250.
- Rack power: GPU racks can need 10 to 20 times the power of older racks.
- Scale: The same source estimates around 12,000 data centers worldwide, with roughly 10% focused on AI.
More power means more heat. More heat means new cooling. That is where the other problems start.
How much electricity do AI data centers use?
A lot, and the number is climbing fast. The International Energy Agency says data center electricity demand rose 17% in 2025. That is more than five times faster than overall global demand growth of 3%.
The IEA also expects data center demand to double by 2030, with AI-linked demand tripling.
Why so much? Two reasons:
- Training a model keeps thousands of chips busy for weeks.
- Running it for millions of daily users never stops.
How much water do they use?
Heat has to go somewhere. Many sites use water to carry it away. Rystad Energy estimates data centers used about 222 billion liters for cooling worldwide in 2025. Without better methods, that could nearly triple to 644 billion liters by 2030.
Tech firms say newer “closed-loop” cooling helps. The coolant circulates inside sealed pipes instead of evaporating. Nvidia says some designs can nearly eliminate water use.
But there are fair counterpoints:
- Sealed cooling often needs more electricity.
- Water used to make power and chips can be about twice a site’s direct use in the US, according to critics cited by the Kuwait Times report.
- Reporting is not consistent across the industry, so claims are hard to compare.
So “less thirsty” does not always mean “lower impact.”
Who is paying for all this?
The money is enormous. Dell’Oro Group raised its 2026 global data center spending outlook to more than $1 trillion. Spending by the four biggest US cloud firms grew 78%, and Dell’Oro expects growth to speed up further in the second half of 2026.
Most of that goes to AI chips, servers, and the buildings around them.
What about the UK?
This is the part most global guides skip. Britain wants to be an AI hub. The government has created AI Growth Zones, areas where it supports large data center projects with better access.
to power and planning help. The zones include Oxfordshire, the North East, North and South Wales, and Scotland.
Reporting from Resul <s11> t <s11> sense says applicants had to show they could build sites of 500 megawatts or more by 2030, with a realistic way to power them.
Here is the catch: the grid. The same report says grid connection waits can run 8 to 10 years. It also says UK electricity is costlier than anywhere else in Europe. Many sites have big plans but unclear power.
In short, the UK has the ambition. Whether it has the electricity is still an open question.
What should you watch for?
You do not need to run a data center to care. Here are four simple signals:
- Where the power comes from. Renewable deals and new grid links matter more than announcements.
- How it is cooled. Ask if a site uses closed-loop systems and what that does to energy use.
- Whether firms publish real numbers. Honest water and power reporting is a good sign.
- Local impact. Jobs and investment are promised. Check what happens to local power and water too.
Frequently asked questions
What is an AI data center in simple words?
It is a facility built to run AI. It holds thousands of powerful chips plus the power and cooling they Need.
Why do AI data centers use so much power?
GPUs draw far more electricity than regular chips, and they run constantly for training and daily use.
Do AI data centers use a lot of water?
Many do, mainly for cooling. Newer sealed systems reduce direct use, but they can raise electricity needs.
Are AI data centers bad for the environment?
They add pressure on power grids and water. The impact depends on the energy source, cooling method, and location.
Final thoughts
AI data centers are the physical engine behind the AI boom. They are powerful, costly, and hungry for energy. The big question for the next few years is simple: can we build them without straining the grid and the water supply?
Next step: Next time you read an AI data center announcement, ask where the power will come from.
