Researchers at UMass Amherst are tackling one of AI's biggest environmental challenges by designing computing systems that bend to match clean energy availability rather than forcing grids to power them constantly. Instead of treating AI infrastructure as a fixed demand, scientists are creating algorithms that pause, delay, or accelerate computing tasks based on real-time signals from the electrical grid, allowing data centers to run primarily on solar and wind power.

Why Is AI's Energy Demand Becoming a Climate Crisis?

Artificial intelligence is consuming electricity at unprecedented rates. U.S. power demand is expected to increase up to 3.5 percent annually through 2040, with data centers representing an increasing share of that growth. The scale of this demand is staggering; as one leading researcher noted, "the scale of the demand and the complexity of AI infrastructure is an order of magnitude greater than it was for previous generations of digital infrastructure".

This explosive growth creates a fundamental problem: powering massive data centers with renewable energy alone is unrealistic under current infrastructure designs. Most data centers rely on hundreds of servers in a single location, requiring constant power regardless of whether the sun is shining or wind is blowing. The mismatch between AI's relentless energy appetite and renewable energy's intermittent nature threatens to undermine climate goals even as clean energy expands.

How Are Researchers Redesigning AI Infrastructure for Renewable Energy?

UMass Amherst faculty have launched two major initiatives to solve this problem. In 2024, a team led by distinguished professor Prashant Shenoy received a $12 million, five-year National Science Foundation Expeditions grant to develop the field of computational decarbonization. The research spans multiple institutions, including the University of Chicago, UCLA, MIT, Carnegie Mellon University, and the University of Wisconsin, and focuses on automating decarbonization across the electrical grid, buildings, transportation, and computing itself.

Additionally, in spring 2026, UMass awarded a Strategic Partnerships to Advance Research and Creative activity grant to establish a large-scale research and workforce development initiative in sustainable computing and AI. Both projects share a common focus: making AI systems flexible enough to work with clean energy rather than against it.

The core innovation involves time-shifting, a technique where algorithms account for grid conditions such as cost, supply-demand balance, and renewable energy availability to optimize when AI workloads run. Rather than processing data immediately, systems can wait for moments when the grid has abundant solar or wind power, then execute tasks in batches. This approach has proven effective enough that Shenoy won an ACM e-Energy 2024 Test of Time Award for his experimental research on the concept.

What Practical Steps Are Researchers Taking to Test These Ideas?

• Solar-Powered Micro Data Centers: Shenoy and David Irwin, a professor in UMass's Riccio College of Engineering, built a micro data center powered by solar panels with battery support at the Massachusetts Green High Performance Computing Center. This testbed allows researchers to experimentally validate carbon-aware algorithms under real-world conditions rather than just in simulations.
• Distributed "Micro Edge" Infrastructure: Rather than concentrating servers in massive centralized data centers, researchers are developing strategies to deploy small collections of servers, called "micro edges," that can be easily powered by renewable sources like solar and wind. While individual micro edges may be less reliable than large fortified data centers, clever algorithms can ensure the overall system remains dependable even if 10 to 20 percent of micro edges are temporarily offline.
• Carbon Intensity Forecasting: Distinguished professor Ramesh Sitaraman developed a tool called CarbonCast that predicts the carbon intensity of a power source up to 96 hours in advance, similar to a weather forecast. This enables algorithms to move computational workloads from locations with fewer renewable sources to those with more, reducing overall carbon emissions.

"AI infrastructure has a flexibility that, if utilized, can be perfectly paired with the recent expansion of clean energy," explained Mohammad Hajiesmaili, associate professor in UMass Amherst's Manning College of Information and Computer Sciences. "One of the biggest issues with clean energy is that it's intermittent; it depends on the availability of sun and wind, for example. But AI infrastructure can be designed to adapt to signals from the grid."

Mohammad Hajiesmaili, Associate Professor, UMass Amherst Manning College of Information and Computer Sciences

Are There More Experimental Approaches on the Horizon?

Researchers are exploring even more unconventional solutions. Sitaraman is investigating the possibility of deploying data centers on Low Earth Orbit satellites, which would offer abundant access to solar power and eliminate water consumption for cooling. However, this approach faces significant challenges; LEO satellites orbit Earth roughly once per hour, requiring sophisticated algorithms to move data efficiently between servers in constant motion. Additionally, the carbon emissions from satellite launches, repairs, and replacements must be carefully accounted for and minimized to make the approach truly sustainable.

The broader vision uniting these efforts is clear: AI and clean energy are not inherent enemies but potential partners. "The future internet holds extraordinary promise and extraordinary risk," Sitaraman warned during a spring 2026 Distinguished Faculty Lecture. "As scientists, we bear a special responsibility to shape which of those futures prevails".

These initiatives represent a fundamental shift in how researchers approach AI's environmental footprint. Rather than accepting that AI will demand constant power from whatever sources are available, UMass and its collaborators are redesigning computing infrastructure from the ground up to work with renewable energy's natural rhythms. If successful, these approaches could transform AI from a climate liability into a tool that actually accelerates the transition to clean energy.