Illustration by Tabitha Walters
As generative artificial intelligence continues its colonization of the digital world, from search engines to software, concerns are being raised about its energy consumption: the data centers that train and run genAI models are power hogs. A December U.S. Department of Energy report cites genAI as the main reason data centers’ share of U.S. energy consumption more than doubled from 2018 to 2023—and may more than double again by 2030.
How is this playing out at the U-M? VP for information technology and chief information officer Ravi Pendse says that in the year since the university introduced its own genAI tools, they have used between three to five megawatts of power—though not in Ann Arbor. They run “in Microsoft Azure Cloud.”
The university chose Microsoft, Pendse says, because “they have been zero carbon since 2012, and they have in their roadmap goals to be carbon negative by 2030.” The company is so intent on not burning fossil fuels that it has contracted with companies to reopen a mothballed nuclear reactor and build small modular ones.
But there’s a caveat to its “carbon negative” claim. According to Microsoft’s 2024 Environmental Data Fact Sheet, its greenhouse gas emissions totaled 17.2 million tons, but its carbon offsets were just five million. A footnote clarifies: “The boundary for this carbon neutral goal includes global Scope 1, Scope 2 market-based, and Scope 3 business air travel emissions.”
Scope 1 refers to direct emissions from resources that Microsoft owns; Scope 2 emissions are produced indirectly from the power it takes to run its operations. Scope 3 encompasses emissions from indirect sources, such as their supply chain—and, at 16.6 million tons, accounts for 97 percent of its emissions. The company aims to reduce Scope 1 and 2 market-based emissions to “near zero” by this year—but to reduce Scope 3 emissions by only “more than half” between 2020 and 2030.
“We need to leverage these tools and technologies with our eyes wide open,” Pendse acknowledges. But, he adds, genAI is already making a significant difference in the lives of people around the world.
Related: Chatbots Go to College
One example: GoToCollege, an app being developed by Pendse’s team that will let prospective college students type in basic information and find all the scholarships they may be eligible for in less than ten seconds.
“And that’s not just to go to Michigan—to go to college anywhere,” Pendse says. “We have been able to very quietly work on it, and our intention is to, sometime in the new year, roll this out.
“It will not be perfect on day one,” he emails, “but we will collect feedback and improve it. We hope that it will be a game-changer for some young people out there.”