Author: Morgan Wolfe
Published: 7:10 PM CDT April 14, 2026
Updated: 7:10 PM CDT April 14, 2026
MINNEAPOLIS — As artificial intelligence continues to expand, so do concerns about the energy demands of the data centers that power it.
At the University of Minnesota, researchers are making big strides on a potential solution: a new type of computing technology designed to significantly reduce how much energy AI systems use.
Inside a campus lab, teams are developing what’s known as Computational Random Access Memory, or CRAM — an approach that could change how computers process information.
“When you see the problem, sometimes people just feel worried, concerned,” said Professor Jian-Ping Wang. “I feel this time I can do something.”
Rethinking how computers work
Traditional computers constantly move data between memory and a processor to perform calculations — a process that requires significant energy, especially for AI systems.
CRAM aims to eliminate that step by allowing computers to process data directly where it is stored.
“We don’t need to move the data back and forth,” Wang said. “We can do the computing inside the memory.”
Researchers say cutting out that data movement could dramatically reduce energy consumption.
From lab to startup
Wang’s concept dates back more than two decades. Now, recent advances in materials and device design are accelerating its development.
He recently founded BesiMax, a startup focused on bringing CRAM technology to market. The team is using silicon wafers to build chips that can transform memory into processing units — a key step toward real-world applications.
“The next step is at the chip level,” Wang said. “These chips can be accepted by major players like Amazon and Microsoft to put in their data centers.”
Addressing growing energy demands
The push comes as AI drives a surge in data center use worldwide.
According to the International Energy Agency, a single AI-powered query can use about 10 times more electricity than a standard internet search. That increased demand has raised concerns about energy consumption and water use tied to large-scale data centers.
BesiMax estimates CRAM could reduce AI energy use by nearly 99%.
Looking ahead
Researchers say the goal is to move from lab testing to scalable chip production — and eventually integration into major data centers. Wang says they’re in the demo phase and could have a prototype ready within two years.
What began as an effort to address a growing problem is now advancing toward a potential breakthrough.
“We are doing our best. We feel very proud,” Wang said.