On the outskirts of America’s booming digital economy, where anonymous warehouses hum with the computations that power artificial intelligence (AI), an invisible problem is growing louder: supplying these data centers with the increasingly large volume of energy they require.
With his latest project, Nebraska Engineering researcher Jun Wang is working to create a solution with wide-ranging positive impacts for those businesses, their surrounding communities and the people who live there.
Wang is developing a high-voltage semiconductor power module — known as LincolnPak — that creates overall energy efficiency the moment electricity enters the building. The U.S. Department of Energy, through its Advanced Research Projects Agency-Energy (ARPA-E), recently awarded Wang a three-year, $1.8 million grant to further work on the new technology.
Today’s large AI facilities typically receive power from the grid at medium voltage — often 13.8 kilovolts — before stepping it down to 480-volt alternating current (AC). From there, the electricity passes through multiple additional conversion stages before reaching the chips inside GPUs, which operate at around 1.2 volts direct current (DC). Each step adds cost, bulk and energy loss.
“The more stages you have, the more inefficiencies you introduce,” said Wang, assistant professor of electrical and computer engineering.
LincolnPak reimagines that entry point. Instead of converting 13.8 kilovolts AC down to 480 volts AC and then repeatedly transforming it again, Wang’s team proposes converting it directly to an 800-volt DC distribution bus. From there, only two additional stages are required to reach chip-level voltage. The result: fewer components, lower losses and a simpler architecture aligned with next-generation data center roadmaps.
At the heart of the innovation is a high-voltage semiconductor packaging technology capable of handling more than 30 kilovolts. That capability enables a new converter circuit that was previously impractical. By integrating the most challenging high-voltage elements into a single advanced module, the LincolnPak design reduces the number of devices and capacitors required in the overall system.
The projected impact is significant. Wang estimates the system could reduce the footprint of this power-conversion stage by at least 50 percent. Construction costs drop. Cooling becomes easier because less energy is lost as heat. Operational efficiency improves, lowering total cost of ownership for companies running massive AI facilities.
The ARPA-E grants are typically awarded, Wang said, to projects that reflect a high-risk, high-reward model focused on disruptive energy technologies.
For everyday electricity users, the payoff of the LincolnPak could be subtle but meaningful: more efficient data centers, reduced grid strain and a digital future that consumes less to compute more.
Wang said he and his collaborators aim not only to demonstrate a working prototype but also to move toward commercialization — potentially launching a startup to license and manufacture the modules.
“We know AI isn’t slowing down,” Wang says. “So, our responsibility as engineers is to make sure the energy infrastructure keeps up: efficiently, sustainably and intelligently. If we can cut losses at the front door of the data center, that impact scales everywhere.”
