Ensuring academic programs keep pace with advancing AI has sparked a costly tech challenge: the need for significantly more campus computing power.
For decades, colleges and universities supported most research needs with separate departmental servers and periodic upgrades. Today, AI and advanced analytics are changing that equation.
Training large language models, processing massive datasets and conducting modern scientific research require high-performance computing that more than quadruples the energy most institutions currently provide—and which they were never designed to support.
Your next read: Here are 4 big ways colleges are reinventing themselves
“The scale required to solve the problems that we have today is just so much more massive than most folks have,” says Steve Goodman, senior director of technology at Marquette University. “This kind of hardware is a must now if you expect to produce anything.”
Campus executives are weighing investments in advanced computing as AI-powered labs and classrooms become the standard for a competitive institution.
Why hardware demand has skyrocketed
High-performance computing is rapidly becoming a prerequisite for new frontiers of academia, Goodman says.
Researchers in medicine, engineering and multidisciplinary fields are tackling high-speed, complex computational workloads. The demand extends far beyond STEM as, for example, philosophy faculty wrestle with the ethical implications of powerful new technologies.
Some universities lack the computing power needed to tackle today’s most complex technological challenges. At Marquette University, a student-built teaching assistant chatbot spent more than two months learning to evaluate coding assignments, only for researchers to discover that its assessments were inaccurate.
The experience illustrates the cost of limited computing resources, says Goodman. “How long is it going to take to find a possible solution to, say, cancer if you have to wait that long to find out that something didn’t work?” he asks.
The problem cascades into enrollment. Prospective students—and their future employers—want a curriculum that covers their discipline’s AI skills.
“Folks will expect to go to a university and learn about AI,” Goodman says. “It’s going to become a differentiator very quickly.”
Purely cloud-based environments will not support most institutions’ computing needs. Continuous, large-scale workloads become too costly and require hands-on control, Goodman says.
“At some point there’ll be a wall between those that have and those that don’t,” he adds. “You don’t want to be on the wrong side of that wall.”
Funding the next generation of research
Funding is the No. 1 problem most institutions will face when securing better hardware. The problem is more complex than a one-time purchase.
High-performance data centers require ongoing investments as AI evolves annually. If delayed, an expensive graphics processing unit used to support AI can become obsolete by the time it’s approved and implemented.
Moreover, the power and cooling required to maintain a data center year-round can exceed a million dollars annually, Goodman says.
Some institutions have tapped industry partners to achieve their goals. North Carolina A&T University is expected to build a 10-megawatt center, in partnership with ImpactData and Raeden.
Fisk University, a small HBCU in Tennessee, recently unveiled its plan to construct a 70,000-square-foot data center on the south side of campus. The university is currently seeking a private industry partner to share the center’s $400 million planning estimate.
The center will feature an academic space and “power new and existing curriculum and tech-forward interdisciplinary study,” a school announcement read.
Donors are also showing interest in supporting AI initiatives that position institutions for the future. At the University of Wisconsin-Madison, an anonymous donor recently committed $100 million to establish the College of Computing and Artificial Intelligence.
The investments will support 50 new faculty positions and “building advanced computing infrastructure,” according to the announcement.
Consolidating compute resources
While expensive to launch and maintain, a centralized computing infrastructure can eventually conserve energy costs, Goodman says.
Many universities still struggle with fragmented technology environments that create redundant requests for computing resources. A shared system could serve multiple research groups simultaneously. Perhaps more importantly, it can encourage different departments to collaborate.
“Hardware that serves one siloed project is a cost,” Goodman says. “Hardware that becomes an institution’s connective tissue is something else entirely.”
