Asking a question to a chatbot or playing a video is just a click on the screen. But behind the scenes, huge data centers are operating every minute of the day. With the rise of artificial intelligence, their energy needs have become one of the most important issues in technological development.
The world's data centers used about 485 terawatt hours of electricity in 2025. This is about ten times the total annual electricity consumption of Hungary. The rate of growth is even more telling: data center consumption increased by about 17 percent in a single year, and that of facilities specifically built on artificial intelligence by about 50 percent. Meanwhile, global electricity demand increased by only 3 percent. According to forecasts, data center consumption could double by 2030, reaching about 3 percent of total global electricity use.
A single large data center can use as much electricity as 100,000 homes. State-of-the-art AI servers are particularly energy-hungry. A single refrigerator-sized server rack can use as much power as 65 homes combined in a few years. Where many of these facilities are concentrated, this can have a negative impact on the national energy system. In Ireland, for example, data centers account for more than a fifth of total electricity consumption.
The majority of the energy is used by the servers and computing chips themselves, but cooling is also a significant item. Every kilowatt-hour consumed is eventually converted into heat, which must be removed or the equipment will overheat. Cooling accounts for less than a tenth of the total consumption in a modern, efficient facility. In an older, less efficient data center, however, it can be as much as a third. Traditional air cooling is increasingly insufficient with today's densely installed AI chips. That's why liquid cooling, which directs coolant directly to the processors, is becoming more common. In some systems, servers are completely immersed in a special, non-conductive liquid.
The good news is that efficiency is improving extremely quickly. The energy demand of an AI task has decreased by up to a tenth per year in recent years, an unprecedented rate in the history of energy. The challenge is that at the same time, usage is growing even faster. More and more people are using AI, and for increasingly complex tasks.
The solution therefore does not lie in a single technology. We need more energy-efficient chips and software, more advanced cooling systems, renewable energy sources and the use of waste heat. A good example of the latter is when heat from a data center is used to heat residential buildings or factories. Digital infrastructure and energy are thus becoming increasingly intertwined. The sustainable technological development of the future depends on how we can design the two together.
