AI can put data centers on an energy diet with smart hardware

Two men at a bench with lots of equipment. Both are standing and looking at the camera. The one in front is wearing a black and red hoodie. The one in the back - a white sweatshirt
As electricity moves from power plants to data centers, much of it is lost through inefficient power electronics. Assistant Professor Feng Guo (back) and doctoral student Knapoj Chaimanekorn, electrical engineering, are improving semiconductor switching, making power electronics faster and much more efficient.

Data centers have a large appetite for electricity – and a bad habit of wasting it. Surprisingly, AI – the very thing that data centers power – could also provide the energy diet they need.

With electricity demand climbing, Feng Guo, assistant professor of electrical engineering, said better energy efficiency could yield significant cost savings.

Guo specializes in developing high-efficiency and high-power-density devices that manage energy performance – equipment that can also be used in electrified transportation systems.

Our students and faculty are making waves of impact.
Here, they are using AI to make and conduct smart devices to save energy on a nationwide scale.

Electricity usually arrives from the grid as alternating current (AC) over transmission lines. But inside a data center, nearly all the equipment runs on direct current (DC). Before it reaches the servers, the power must pass through a host of devices like transformers, converters and controllers, stepping through multiple stages to change the current or the voltage.

Along the way, each conversion leaks energy.

Devices guide the electricity journey

Two men at a lab bench looking at the camera. The one in the back is wearing a dark hoodie. The one in the front is wearing a white UWM sweatshirt and is holding a circuit board.
Smart hardware can dramatically improve energy efficiency. Guo (front) holds a component of semiconductor testing, while Chaimanekorn displays the paralleled power electronics converters which increase the output current and power.

Data centers are raising their DC voltage to pack more power into the same space and boost efficiency. This requires redesigning equipment, collectively called power electronics, needed to shepherd the flow of electricity in ways that achieve that.

Researchers feed AI models vast amounts of electrical data, from voltage and current to frequency patterns. From the data, AI develops a deep understanding of how these systems behave. That knowledge informs smarter designs for converters and other crucial components that waste less energy from the start.

And that’s only the beginning. After engineers model the system, then AI can figure out the most efficient playbook.

Think of AI as a conductor, orchestrating “smart” components that generate, route, and control power. It can even take on tricky jobs like fault detection and managing renewables.

“Each time you train the model, you’re asking a slightly different question,” Guo said. “It’s a process that strengthens the model. You start with the ocean, then zoom in, little by little.”