How Much Will AI Drive Electricity Demand?

The data centres and processors that train AI are consuming more and more electricity every month. But how significant really is AI in the context of all energy consuming activities and industries? 

How much is AI driving electricity demand?

The queries you send to an AI chatbot and the responses you receive back are relatively low-resolution. That’s compared to Zoom calls or streaming video, for example.

At the AI’s host server, however, calculations are being made involving billions of numerical parameters. Reportedly, over 2.5 billion requests are made to OpenAI, the host of ChatGPT, every day.

Those calculations take up a tremendous amount of computing power. Computation requires energy, it generates heat, and the heat requires cooling that demands even more energy.

The computer servers being used to train AI are mounted in standard 19-inch racks. These are vertical metal cabinet that are 19 inches (48cm) wide and 1.5m to 2m tall.

In AI data centres, these racks host an array of the most densely constructed and advanced computer chips. Immersed in liquid coolant and fully firing, a single one of these racks in 2025 consumes an astonishing 165kW of energy. That’s around 10% of the power produced by an entire, fully operational coal power station.

Source

Data centres in their entirety, not just those used for AI, were estimated to be consuming 28.6GW of energy at the end of 2024. That’s about 18 large coal power stations.

On one estimate, this is set to increase to around 92GW by 2027. This is greater than the current electrical generation capacity of countries like France, Spain and Italy.

It is no surprise, then, that Big Tech companies are investigating ways to generate their own power. Both Microsoft and Google have reportedly invested in next-generation “modular” nuclear power. This would involve fabricating nuclear reactors off-site and then shipping and siting them next to their data centres.

Forecasts suggest data centres will consume 4% of global energy demand by 2030. In the US, they are forecast to consume 8% of national demand, according to analysis.

AI data centres are the most energy infrastructures ever invented. How much of our resources they consume will ultimately depend on how long the AI bubble continues to inflate?

Sign Up To Our Free Newsletter