AI feels weightless. You type a question, and an answer appears in seconds. But behind that answer are huge buildings full of servers that need a lot of electricity, both to run the chips and to keep them cool. As AI use explodes, so does its appetite for power.

In June 2026, research firm Gartner estimated that data centers worldwide will use 26% more electricity this year than last year, with AI as the main driver. It even said AI growth is now limited by how much power is available. Here’s what the numbers show, where all that energy goes, and why making computing more efficient is one of the most important challenges for future engineers.

What’s Happening

According to Gartner, global data center electricity use will rise from 447 terawatt-hours (TWh) in 2025 to 565 TWh in 2026. One terawatt-hour is a billion kilowatt-hours, enough to power around 100,000 typical US homes for a whole year. Total data center power demand is expected to grow from 104 gigawatts to 132 gigawatts in the same period.

AI is the main reason. Servers designed for AI are expected to use about 31% of all data center electricity in 2026, up from around 20% in 2025, and their consumption is growing by over 80% a year. Meanwhile, ordinary servers are barely growing at all. Gartner expects AI servers to use more electricity than conventional servers by 2027. Cooling systems are growing fast too, by 22.6% in 2026.

Looking further ahead, Gartner projects data centers could use more than 1,200 TWh by 2030, roughly the same as Japan’s entire annual electricity demand. “AI capacity is now constrained by power availability,” said Gartner’s Linglan Wang.

The Tech Behind It: Where the Power Goes

Most of a data center’s electricity goes into two things: computing and cooling. AI chips like GPUs perform enormous numbers of calculations in parallel, and modern AI servers can pack many of these chips into a single rack. All that work turns into heat, so the more powerful the chips, the more cooling they need.

Traditional data centers cool their servers with air, using large fans and air conditioning. But AI racks produce so much heat that many new facilities are switching to liquid cooling, where cool liquid flows through plates attached directly to the chips, carrying heat away much more efficiently. Engineers also track a measure called Power Usage Effectiveness (PUE), which compares the total power a data center uses with the power that actually reaches the computers. The closer it is to 1.0, the less energy is wasted.

Efficiency also happens inside the chips and software. Newer chips, like those built on TSMC’s 2nm process, do more work per watt. Smarter software can schedule tasks, shrink AI models, or run them on more efficient hardware like NPUs. Every small improvement, multiplied across millions of servers, saves a huge amount of energy.

What It Means for Indonesia and Future Engineers

This topic hits close to home. Indonesia’s data center capacity is around 800 megawatts and could reach 1.3 gigawatts with projects in the pipeline, and the government is promoting places like Batam as data center hubs, pointing to land, electricity, green energy, and water as key advantages. As those projects grow, the question of how to power and cool them efficiently, especially in a hot, humid climate, becomes a real engineering challenge.

That creates demand for engineers who understand energy as well as computing. Some will design more efficient chips and servers. Others will build monitoring systems with sensors and software that track power and temperature in real time, or develop control systems that connect data centers to solar and other renewable energy. Even software engineers play a role by writing code that uses less computing power.

In short, the future of AI doesn’t only depend on smarter algorithms. It depends on engineers who can make computing faster, cooler, and greener at the same time.

Key Takeaways

Gartner expects global data center electricity use to jump 26% in 2026, from 447 TWh to 565 TWh, with AI servers using about 31% of that power. By 2030, data centers could use over 1,200 TWh a year, about as much as Japan. Better chips, liquid cooling, smarter software, and renewable energy are all part of the solution, and each one needs engineers.

Want to build technology that’s both powerful and sustainable? The Computer Engineering program at BINUS University teaches you how computers, embedded systems, and energy-efficient hardware work. Explore the program and help shape a greener future for AI.

Sources: