AI data centers consume 6% of electricity in the UK and US

Data centers powering artificial intelligence and cloud services consume 6% of electricity in the UK and US. This figure is growing rapidly and already causing public resistance in local communities. New research shows that the energy cost of AI will become one of the key issues of the next decade.

Investments in data centers at the level of state economy

The International Data Center Association (IDCA) published research showing that annual global investments in data centers are approaching $1 trillion. This is approximately 1% of the world economy — a huge share comparable to the economies of entire countries. The figure demonstrates how critical data storage and processing infrastructure has become in the modern world.

Electricity consumption by data centers has grown by 15% globally over the past two years. This growth rate is more than twice the average growth in electricity consumption and is primarily driven by the acceleration of AI system implementation — from training large language models to launching applications in production.

To understand the scale: if five years ago video streaming and cloud storage were the main consumers of energy in data centers, today AI has been added, which requires orders of magnitude more power.

• Training one large neural network requires several months of continuous work on powerful GPU clusters
• Running inference (calculating answers for users) requires a constant stream of computations 24/7
• Duplication and redundancy of infrastructure for reliability increases total consumption by 20-30%
• Server cooling becomes one of the largest expense items and can reach 40% of total energy consumption

Why AI requires so much electricity

Behind every chatbot you use are massive computational processes. While data centers previously powered mainly websites, streaming services, and social networks, AI has now been added, requiring orders of magnitude more power.

Training a single large model can require as much electricity as thousands of homes consume in a year. For example, training GPT-4 required approximately as much energy as a small city consumes in a month. And when the model is ready, its operation for millions of users requires a constant stream of energy, because servers work without breaks.

Additionally, companies often train multiple versions of one model and keep them in memory for quick access. This further increases energy costs.

"The energy cost of AI is a hidden cheat code in the economy," noted one of the IDCA researchers in an interview with

The Guardian.

Local communities rise up against data center expansion

The wave of AI development has led to conflicts with local communities. Residents oppose the construction of new data centers, fearing rising electricity bills and degraded quality of life. In some areas of the UK and US, resolutions against capacity expansion have already been adopted, and in some cases, permissions for new construction have been frozen.

The problem is acute: if data centers continue to consume electricity at the current growth rate of 15% per year, it will seriously compete with the needs of ordinary citizens, hospitals, and industry. In some regions, energy shortages during peak hours are already being observed.

What this means

The AI developer community and companies like OpenAI, Google, Meta have faced harsh reality: technological progress has a real energy cost. Either large-scale transition to renewable energy sources (wind, solar, nuclear) will be needed, or the algorithms themselves will need to be optimized to require fewer computations. Both paths require enormous investments and time.

Until this happens, the energy crisis will be one of the main limiting factors in AI development.