Minnesota startup has found a way to use lasers to reduce energy and water use by data centers

Big technology companies have committed hundred of billions of dollars to build out data centers, mainly to process AI.

The data centers represent economic opportunity for communities, but they are also meeting with more vocal resistance from citizens. More people have become concerned about how much power and water these data centers will need to operate.

Companies from 3M to nVent and Ecolab have products and solutions to keep data centers cool, a big problem when you have a building full of machines working 24 hours.

A startup called Maxwell Labs in Little Canada is developing a technology that would cool down the chips using lasers instead of water.

“Three major problems are facing data centers and AI,” said Mike Karpe, the startup’s co-founder and chief revenue officer. ”Energy, heat and performance. And you know this is really going to be coming to the head in the next five years."

Maxwell Labs says the laser technology will save energy and unlock more computing power from today’s ultrahot and fast-running AI computer chips.

Karpe founded Maxwell Labs in 2020 with Jacob Balma and Alejandro Rodriguez.

“Next step is basically scaling our team, developing the capabilities in the lab and then focusing on this demonstration,” Rodriguez said.

The company has found backers who believe in the technology’s potential.

To date, Maxwell Labs has raised $4.5 million from institutional and angel investors, including $500,000 from a joint program of the U.S. Army Combat Capabilities Development Command Army Research Laboratory, the University of St. Thomas and the Energetics Technology Center Inc.

Now, the company is working on a two-part Series A venture capital round that could land $5 million by January and $25 million by the end of 2026.

About 6% of the U.S. energy supply goes to serving data centers. The large-scale data center buildouts being proposed right now could double that by 2028.

Air and liquid are types of conductive cooling. But Karpe describes that switch as a Band-Aid approach.

The conductive methods cool the entire chip. They can’t target the roving micro hotspots in the precise way lasers can.

Maxwell Labs’ approach depends on photonics, the study of ways to control wavelengths of light at micrometer and nanometer scale through integration and miniaturization of optical components.

In addition to partnering with Rodriguez and his Princeton team, Maxwell Labs is also collaborating with Sandia National Laboratories and the University of New Mexico on the materials that will best work in the system.

For decades, photonics has contributed to advancements in data transfer, solar cells and LED lights. Only recently has the science explored the right conditions where lasers can be used for cooling.

That power density translates to heat, and it has become a limiting factor in computer chip performance. And fast- and hot-running artificial intelligence chips have created the huge need for power and water to cool the chips.

As a computer chip works, it doesn’t heat up uniformly. Depending on what the chip is asked to do, individual transistors on the chip work and heat up in different areas and different levels, creating micro hotspots.

Using advanced optics and lasers, Maxwell aims to target those roving hotspots and convert that heat to light, recycle that light and turn it back into electricity.

Balma says once you convert the heat to light, you can recover that energy.

“The types of photovoltaics that you use in this scenario are called laser power converters,” Balma said. “And they’re way more efficient than solar panels, so instead of like, 20 percent or 30 percent efficiency, they’re like 80 percent to 90 percent.”

It is economical, too.

“It has the possibility of completely covering the cooling cost of data centers with the waste heat recovered,” Balma said.

Maxwell Labs believes the key to unlocking even more computer chip capabilities lies in cooling chips more efficiently using radiant properties within photonics as well as conductive cooling.

“Long term, if you want to make chips as fast as physically possible, you have to also start using radiation to get rid of the heat,” Balma said.

Maxwell Labs’ technology is a long ways from working at a hyperscaler or data center level. But Karpe and Balma believe they soon could be working with niche manufacturers or developers seeking high returns on investments.

Newman retired a few years ago and promised his wife he wouldn’t start a new project unless it was for something where he could give back to the industry.

When he met the Maxwell Labs partners in 2023, he became convinced of the technology and became an adviser and investor that December.

Newman credits Balma, Karpe and Rodriguez for staying on track in Maxwell’s development. He’s encouraging them to finish the prototype and show it to the largest chip manufacturers.

He believes that once one large chip manufacturer signs on others will quickly follow.

“They’ve made a hell of a lot of progress in the last two years,” Newman said. “It became obvious to me that this could be humanity changing.”

Minnesota’s history with supercomputers from companies like Control Data and Cray and local chipmakers Skywater and Polar Semiconductor, Newman and others believe there is a lot of tech talent in the area to draw from.

And the problem Maxwell Labs is trying to solve could become more urgent.

“It’s not like we’re going to be asking less of our of our compute capacity,” said Reed Robinson, a partner with Minneapolis-based Groove Capital and among the first institutional investors in Maxwell Labs.