AI data centres use power not only to run computers, but also to remove the heat produced by those computers.
Introduction
AI feels like software, but behind every AI answer there is real hardware using real electricity.
Every time an AI model is trained or used, thousands of computer chips process data. These chips consume electrical power. A large part of that electrical energy eventually becomes heat. That heat must be removed continuously, otherwise the equipment can overheat.
So the power demand of an AI data centre is not just about computers. It is also about cooling, backup systems, power conversion, networking, and infrastructure.
What is electrical power?
Power tells us how fast energy is being used.
Simple formula:
Power = Energy ÷ Time
Unit:
1 watt = 1 joule per second
So a 1000-watt device uses 1000 joules of energy every second.
For homes we usually talk in watts or kilowatts. For data centres, the numbers become huge: megawatts.
Why AI chips need so much power
AI workloads need many calculations.
These calculations happen inside GPUs, CPUs, memory systems, and networking equipment. When many chips run together, the total power demand becomes very large.
A single chip may use hundreds or even thousands of watts. A rack of servers can use many kilowatts. A full AI data centre can need tens or hundreds of megawatts.
Where does the energy go?
This is the key Physics point:
Most of the electrical energy used by computers eventually becomes heat.
The chips do useful computational work, but the energy does not disappear. It mostly ends up as thermal energy.
An AI data centre does not just host computing machines. It is also a giant heat-producing system.
Why cooling becomes a big challenge
If heat is produced continuously, it must be removed continuously.
Cooling systems may include:
- Air conditioning
- Fans
- Chilled water systems
- Liquid cooling
- Cooling towers
- Pumps and heat exchangers
All of these also need electricity.
So more computing power means more heat, and more heat means more cooling power.
What is PUE?
PUE means Power Usage Effectiveness.
Formula:
PUE = Total data centre power ÷ IT equipment power
Example:
If the computers use 100 MW and the whole facility uses 130 MW:
PUE = 130 ÷ 100 = 1.3
A perfect data centre would have PUE = 1.0, but in reality extra power is needed for cooling, lighting, power systems, and other infrastructure.
Why demand can grow fast
AI demand can grow because of:
- Larger models
- More users
- More training runs
- More AI features in apps
- More data storage and movement
- More powerful chips packed into smaller spaces
That means the power requirement can rise quickly, especially when many organisations start using AI at scale.
The Simple Physics Summary
AI data centre power demand is high because:
AI needs many calculations.
Calculations need electrical energy.
Electrical energy mostly becomes heat.
Heat must be removed.
Cooling also needs power.
So the total power requirement becomes much larger than just the computers.
Why This Matters
AI may look like a cloud service, but physically it is powered by electricity, limited by heat, and managed through cooling. That is why the future of AI is also a Physics problem.
Next Blog
Watch out for the next article in this series, where we look at the Physics of Cooling Data Centres.