Universities are tackling AI's energy crisis by designing systems that work with renewable power instead of against it. Rather than simply trying to use less electricity, researchers at UMass Amherst and partner institutions are developing computational approaches that align AI infrastructure with when clean energy is available, turning a fundamental mismatch into a strategic advantage.

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

Artificial intelligence is reshaping how much electricity the tech industry consumes. Google's own environmental report revealed that its greenhouse gas emissions rose by 48 percent between 2019 and 2024, primarily due to increased data center energy consumption and supply chain emissions. The scale of the problem is staggering: U.S. power demand is expected to increase up to 3.5 percent annually through 2040, with data centers representing an ever-growing share of that demand.

"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," said Mohammad Hajiesmaili, associate professor in UMass Amherst's Manning College of Information and Computer Sciences.

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

The challenge is not just about raw energy consumption. Data centers require massive amounts of water for cooling, and they often operate in locations where renewable energy sources are limited or unpredictable. This has created a fundamental tension: as AI becomes more powerful and accessible, it demands more electricity at precisely the times when the grid may be relying on fossil fuels.

How Can AI Systems Be Designed to Match Renewable Energy Patterns?

The breakthrough insight from UMass researchers is that AI infrastructure has a unique flexibility that previous computing systems lacked. Unlike traditional data centers that must run continuously at full capacity, AI workloads can be shifted in time without compromising performance. This opens the door to a radical rethinking of how computing and energy infrastructure interact.

"AI infrastructure has a flexibility that, if utilized, can be perfectly paired with the recent expansion of clean energy," explained Mohammad Hajiesmaili. "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

Prashant Shenoy, a distinguished professor of computer science at UMass Amherst, has been pioneering this approach through experimental research on time-shifting energy usage. His work focuses on developing algorithms that account for real-time grid conditions, such as cost, supply-demand balance, and the availability of green energy, to optimize when AI workloads run. Rather than forcing renewable energy sources to match computing demand, the new approach shifts computing demand to match renewable energy availability.

A key innovation in this research is CarbonCast, a tool developed by Ramesh Sitaraman, a distinguished professor of computer science at UMass Amherst. Similar to a weather forecast, CarbonCast predicts the carbon intensity of a given power source over the next 96 hours, allowing algorithms to move computational workload from locations with fewer renewable energy sources to those with more abundant clean power.

Steps to Implementing Carbon-Aware Computing Infrastructure

• Real-Time Grid Monitoring: Deploy systems that continuously track grid conditions, including the current mix of renewable versus fossil fuel energy sources, electricity prices, and supply-demand balance across different regions.
• Workload Scheduling Algorithms: Develop computational methods that delay non-urgent AI tasks until periods when renewable energy is abundant, while prioritizing time-sensitive workloads during those windows.
• Distributed Micro Data Centers: Build small collections of servers powered by local renewable sources like solar and wind, rather than relying exclusively on massive centralized data centers that require constant fossil fuel backup power.
• Experimental Testbeds: Create real-world testing environments, such as solar and battery-powered micro data centers, to validate carbon-aware algorithms under actual grid conditions before large-scale deployment.

To validate these concepts, Shenoy and David Irwin, a professor in UMass's Riccio College of Engineering, built a micro data center powered by solar with battery support at the Massachusetts Green High Performance Computing Center. This testbed has enabled them to experimentally test carbon-aware algorithms under real-world conditions, proving that the approach works in practice.

What Funding and Research Initiatives Are Driving This Work?

The scale of institutional commitment to this challenge is significant. In 2024, UMass Amherst received a $12 million, five-year National Science Foundation Expeditions grant to develop the field of computational decarbonization. The research team, led by Shenoy, is working with collaborators at the University of Chicago, UCLA, MIT, Carnegie Mellon University, and the University of Wisconsin to apply computational approaches across multiple sectors, including the electrical grid, the built environment, transportation, and computing itself.

Additionally, in spring 2026, UMass awarded a Strategic Partnerships to Advance Research and Creative activity (SPARC) grant to a group of researchers led by Hajiesmaili, with the goal of establishing a large-scale research and workforce development initiative in sustainable computing and AI. These investments reflect a growing recognition that solving AI's energy problem requires not just incremental improvements, but fundamental rethinking of how computing infrastructure is designed and operated.

The research also extends to more experimental approaches. Sitaraman is exploring the possibility of deploying data centers on Low Earth Orbit satellites, which would have abundant access to solar power and require no water for cooling. However, this approach presents unique challenges, including the need for smart algorithms to move data between servers in constant motion and accounting for the carbon emissions from satellite launches and repairs.

Are Green Search Engines Part of the Solution?

While universities focus on infrastructure-level solutions, some companies are taking a different approach to address AI's environmental impact. Green search engines like Ecosia and Lilo have emerged as alternatives to Google and Microsoft's Bing, offsetting the environmental impact of online searches by donating a portion of their revenue to environmental causes.

Ecosia, based in Berlin, Germany, uses search results from Microsoft Bing, Google, and the European Search Perspective, depending on location and device type. The platform donates profits from its advertising to environmental organizations and has even built its own solar plants to produce twice the energy needed to power searches. Similarly, Lilo, based in Paris, France, redistributes 80 percent of its advertising profits to environmental projects chosen by users.

However, these green search engines remain metasearch engines, meaning they still rely on larger search engines and their underlying data centers to generate results. While they offset environmental impact through donations, they do not fundamentally solve the energy consumption problem at the infrastructure level. The more comprehensive solution lies in the research being conducted at universities, which aims to redesign computing systems themselves to be inherently more sustainable.

The convergence of these efforts, from algorithmic innovation to infrastructure redesign to alternative business models, suggests that the tech industry is beginning to grapple seriously with AI's environmental footprint. As Ramesh Sitaraman warned in his spring 2026 Distinguished Faculty Lecture, "The future internet holds extraordinary promise and extraordinary risk. As scientists, we bear a special responsibility to shape which of those futures prevails".