Major tech companies face critical power consumption challenges from AI model training and inference, forcing urgent infrastructure adaptations.

The artificial intelligence industry faces an urgent energy crisis as data center consumption threatens global electricity supplies. According to the International Energy Agency, data centers could consume three percent of the world's electricity by 2030, double what they use today. Experts at McKinsey warn that the world is heading toward an electricity shortage while racing to build enough data centers to support AI's rapid growth. The challenge, as Mosharaf Chowdhury, a University of Michigan professor of computer science, explains, can be addressed through two main approaches: either "build more energy supply – which takes time and the AI giants are already scouring the globe to do – or figure out how to consume less energy for the same computing power." Chowdhury believes the challenge can be met with "clever" solutions at every level, from physical hardware to AI software itself.

Significant progress has already been made in reducing operational energy demands. Twenty years ago, operating a data center required as much energy for cooling systems and infrastructure as running the servers themselves. Today, operations use just 10 percent of what the servers consume, according to Gareth Williams from consulting firm Arup, largely through focused efficiency improvements. Many data centers now use AI-powered sensors to control temperature in specific zones rather than cooling entire buildings uniformly, allowing them to optimize water and electricity use in real-time, according to McKinsey's Pankaj Sachdeva. Liquid cooling represents a major breakthrough, replacing air conditioners with coolant that circulates directly through servers. As Williams noted, "All the big players are looking at it."

This innovation is critical because modern AI chips from companies like Nvidia consume 100 times more power than servers did two decades ago. Amazon's AWS recently developed its own liquid cooling method to cool Nvidia GPUs without rebuilding existing data centers. Dave Brown, vice president of compute and machine learning services at AWS, stated in a YouTube video: "There simply wouldn't be enough liquid-cooling capacity to support our scale." Additionally, Mosharaf Chowdhury's lab has developed algorithms that calculate exactly how much electricity each AI chip needs, reducing energy use by 20-30 percent.

While each new generation of computer chips is more energy-efficient than the last, long-term energy consumption will continue rising. Research by Purdue University's Yi Ding has shown that AI chips can last longer without losing performance, yet "it's hard to convince semiconductor companies to make less money" by encouraging customers to maintain existing equipment, Ding added. As Ding predicted, "Energy consumption will keep rising" despite efficiency efforts, "but maybe not as quickly." Energy efficiency has become strategically crucial, with the United States now viewing energy as key to maintaining competitive advantage over China in AI. In January, Chinese startup DeepSeek unveiled an AI model that performed as well as top US systems despite using less powerful chips and less energy, achieving this through more precise GPU programming and skipping an energy-intensive training step previously considered essential. China is also feared to be significantly ahead of the United States in available energy sources, including renewables and nuclear.