Harvard’s Kempner Institute Bet Big on AI. Now It Has to Prove Itself.

Karel Svoboda thought the Kempner Institute might be making a mistake.

As a former member of Kempner’s Scientific Advisory Board, he was skeptical of Harvard’s decision to build its own artificial intelligence computing cluster rather than rely on cloud providers. Owning hardware meant maintenance, staffing, and a long-term commitment to machines that could easily become outdated.

“My opinion was that it may have been a bad idea,” he said. “I was wrong.”

That reversal now sits at the center of Kempner’s first years. Founded with a $500 million gift from Meta CEO Mark E. Zuckerberg and Priscilla Chan ’07, the institute bought computing power before demand for AI hardware surged — giving Harvard researchers a resource many academic labs still struggle to secure.

Svoboda said Zuckerberg pushed Kempner toward building its own computing resources, anticipating that cloud providers would face rising demand as artificial intelligence expanded.

“I actually believe that Zuckerberg himself was pushing this,” he said, adding that Zuckerberg felt the institute could accomplish more with their own cluster.

Since Kempner was announced in December 2021, nearly a year before ChatGPT was released to the public, computing power has become one of the clearest divides between academic AI labs and private companies. Kempner has responded by building one of academia’s strongest AI clusters, creating fellowship programs for students and postdocs, and turning the sixth floor of the Science and Engineering Complex into a hub for researchers studying intelligence in both brains and machines.

The Crimson spoke with more than 20 Kempner affiliates, Harvard researchers, and faculty at peer institutions. Many said the institute has quickly changed Harvard’s AI landscape by giving researchers access to resources that remain scarce across much of academia. But several also said Kempner’s early years have exposed the limits of academic AI research: private labs still set…

Source