A building packed with computers where AI is trained and run.
A data center is a building filled with thousands of computers called servers, plus GPUs. When you ask an AI a question, it usually travels to a building like this, gets processed and comes back as an answer. The machines run nonstop and get hot, so keeping them cool also takes lots of electricity and water.
AI has set off a race to build more of them. The International Energy Agency estimated that data centers used about 1.5% of the world's electricity in 2024, and that their use could more than double by 2030. Where all that power will come from has become a big question.
Think of it like this
Think of a school cafeteria. Instead of every family cooking separately, one big kitchen makes hundreds of meals at once. A data center is a giant cafeteria for AI calculations, and it needs a lot of power and water to keep going.
For example
Google measured in 2025 that a typical text question to Gemini used about 0.24 watt-hours, less than watching TV for nine seconds.
Common mix-ups
Mix-upAI answers are made inside my phone.
ActuallyTurn on aeroplane mode and ask a chat AI app something, and most of the time no answer comes back. That's because the question has to travel over the internet to a distant building and back; only a few light jobs, like dictation, finish inside the phone.
Mix-upData centres only need a little electricity.
ActuallyBy its own account, xAI fitted 100,000 GPUs into its Colossus site in Memphis in 122 days in 2024, but at first the grid supplied only 8 MW, so it had to lean on temporary gas generators. Getting the electricity in was slower than installing the chips.
Try it in class
For teachers at the start of a lesson — or for students working on their own.
- Question
- What everyday thing could you compare the electricity of one AI question with?
- Activity
- Using the measurement in the example above (a median of about 0.24 watt-hours per text question), multiply by the number of questions your class asks in a day. Compare the total with familiar things like charging a phone or leaving a TV on, in a table. Then underline the size figures in the Colossus (E0427) and Stargate (E0498) records and compare them.
Records about this
Source-checked moments from AI history.
Officially acknowledged that power and grid connection delays are the physical bottleneck limiting the speed of near-term expansion
The symbolic starting point of the private AI infrastructure capex cycle
The fastest build-out in the industry, by the company's own announcement. But the initial grid connection was only 8 MW, so it relied on temporary gas turbines, and was stepped up to 150 MW only in November