Ever wondered how much juice your favorite AI apps drink? A new report from the Beijing-based Global Energy Interconnection Development and Cooperation Organization (GEIDCO) at the 2025 Global Energy Interconnection Conference reveals some eye-opening numbers. 💡⚡
Since 2010, the capacity of accelerated computing servers handling AI tasks has grown four times faster than all servers combined. By 2024, global data centers consumed about 415 billion kilowatt-hours—roughly 1.5% of the world’s electricity—a fourfold jump since 2005.
Between 2020 and 2024 alone, electricity use for accelerated computing servers soared 2.4 times, with an average annual growth rate of 36%. That’s like adding a new streaming platform every month! 🎬🔥
But it’s not all power-hungry drama. Smarter software, advanced hardware, and AI-based energy management are turning the tide. Today’s cutting-edge systems can cut about 15% off the energy used for cooling data centers. Pretty cool, right? ❄️
According to GEIDCO, AI power demand will first rise in a rapid, almost linear way—and then, thanks to technological progress and efficiency gains, it will follow a logarithmic curve: super-fast growth at the start, then a gradual slow-down until it levels off. Think of sprinting out of the gates and then settling into a steady jog. 🏃♂️💨
As intelligent applications spread, AI-driven computing will take over more general-purpose tasks, becoming the main power consumer in data centers. Yet, even at its peak, AI’s overall share of global electricity consumption will stay relatively small.
Bottom line: AI is on track to be a key driver of regional electricity growth and will challenge grids to keep pace—but it won’t send the world’s power bills into the stratosphere. 🌍🔌
Reference(s):
Report shows global AI power demand to rise rapidly, then stabilize
cgtn.com
