Universities have long grumbled about losing faculty to industry. The artificial-intelligence boom has turned a tolerable leak into a raid. The recruiters come bearing things no bursar can match: vast computing power, proprietary data and, with share offerings reportedly in prospect for the biggest labs, the possibility of genuine wealth.
Sarah Soule, dean of Stanford University’s Graduate School of Business, admits as much. “There is an allure for faculty,” she told Business Insider. “I won’t lie, I worry about this.” The evidence sits down the corridor: two of her professors, Andrew Hall and Chad Jones, are both on leave this year. Stanford is far from alone. Ms Soule had just shared a public panel with Aaron Chatterji, OpenAI’s chief economist, who is himself on leave from Duke University.
Stanford is sanguine about such absences, partly because they are rationed. Tenured professors may take up to two years of unpaid leave for private-sector work in every seven; government service earns longer dispensation, as Condoleezza Rice enjoyed when she left her post as the university’s provost to become secretary of state. And professors do come back, the university points out: Westley Hartmann, on the faculty since 2003, returned from seven unpaid months at Amazon.
That concedes the deeper problem: the labs increasingly write the job description of the modern star researcher. They have amassed capital, computing power and market influence in quantities no faculty committee can conjure. A stint inside one lets a scholar help steer a technology with enormous economic and social consequences, and share in its financial upside — though at least one prominent endowment chief warns those riches may prove precarious.
Formally, the university refuses to score the traffic as a loss. Departures and collaborations, it says, belong to a broader ecosystem that keeps professors close to the technological frontier and lets them ferry the experience back to their research, teaching and students. “There’s no question that the opportunity to work with frontier AI labs is attractive to faculty,” says Kristin Harlan, the business school’s head of dean’s communications. “We see those relationships as complementary to Stanford’s academic mission, not separate from it.”
Ms Soule’s own counter-offer is not money but academic freedom. “I suspect that the freedom to do what you want to do and to take a very long time to do it will be possibly something that won’t be the case at the frontier labs,” she says. Where commercial labs live from one model release to the next, a professor can spend a decade on a single question. She is betting that some scholars will always prefer that trade.
A costly arms race
Recruiting is only half the squeeze; the other half is equipping those who stay. Over the summer Stanford entered an enterprise agreement with Anthropic, giving students, faculty and staff access to its Claude suite of tools, and the university also pays for shared computing clusters, cloud capacity and hardware for research. But a chatbot subscription is small beer beside what serious AI work consumes. Ms Soule says her school’s AI-related research costs are rising, and that future spending is “super hard to forecast” — while the frontier labs can lay on precisely the computing power and data that provosts struggle to match.
“I don’t have a good answer for it,” she admits of the budgets. “Probably, most deans and most college presidents are really worried about this issue, because it’s really daunting to think about.”
None of this dooms the universities: they still mint the doctoral graduates the labs compete to hire, and a campus leave is not yet a resignation. But the bargain is shifting. Stanford’s two-years-in-seven rule now looks less like an indulgence than a firewall: grant talent its sabbatical and its equity stake, or watch it quit outright. The institution that sells freedom, meanwhile, is rediscovering a law its own economists could have stated: in any bidding war, sell the thing the other side cannot buy. Time, happily, cannot be purchased by the billion.