The Best AI Researcher in the World Isn't Coming to Your Campus (Yet)

Let me describe a conversation that is happening right now at colleges and universities across the country, probably including yours. A provost, a dean, or a research vice president is sitting across from a candidate—a genuinely brilliant machine learning, computational biology, or data science researcher—making the case for why their institution is the right place to build a career. The candidate is nodding politely and then asks the question every higher education leader dreads: "What does your compute infrastructure look like?"

At most institutions, the honest answer involves a pause that says more than the words that follow. This situation is emblematic of the infrastructure gap in higher education, a divide that is widening faster than most technology leaders have had time to fully reckon with. The artificial intelligence (AI) revolution has not merely raised the bar for research computing; it has moved the bar to a different building, one with the power and infrastructure required to support the computing demands of modern workloads.

Meeting Research Computing Demands

There is a persistent and expensive misunderstanding in higher education about what AI-capable infrastructure means. It is not, generally speaking, a matter of buying a few GPU servers and pointing a researcher at them. The workloads that serious AI research generates, such as large language model (LLM) training, high-resolution imaging analysis, genomic computation, and real-time simulations, demand sustained, high-density compute at a scale that exposes every limitation in an infrastructure designed around pre-AI research computing needs.

I spent a significant portion of my career at GE Medical Systems designing infrastructure frameworks for environments that could not afford to get it wrong. Integrating more than three hundred acquired businesses into a coherent technology architecture taught me, among other things, that the gap between what an organization thinks its infrastructure can handle and what it actually can handle is almost always larger than anyone wants to admit. Colleges and universities are no exception. They are currently experiencing the same challenge.

The physical reality is that a rack of current-generation AI accelerators can draw 60 to 100 kilowatts of power, while traditional academic data centers were designed around racks drawing 5 to 15 kilowatts. The gap is not a software update. It is a facilities project, a power upgrade, and, in many cases, a fundamental rethinking of cooling architecture, ranging from air-based systems that worked fine for two decades to liquid cooling technologies that the building may not have been designed to accommodate.

None of that is insurmountable. But none of it is fast, either. And in a talent market where a researcher who is choosing between two institutions is also choosing between two different versions of their next five years of work, "we're working on it" is not a compelling answer.

Recruiting Research Talent

Higher education institutions have always competed for faculty. What has changed is the negotiation "currency." For a generation of researchers whose methodologies run on advanced computing capacity, infrastructure is not a perk; it is a prerequisite for conducting it. An institution cannot recruit a world-class AI researcher to a campus where the research computing environment will spend half its time as a bottleneck and the other half as a topic of conversation at the faculty senate.

Students follow researchers. Researchers follow infrastructure. The sequence matters, and right now the institutions that understand it are pulling ahead of the ones still treating research computing infrastructure as a line item rather than a strategic asset.^Footnote1

The downstream effects compound quickly. Strong researchers attract strong graduate students. Strong graduate students attract industry partnerships, sponsored research, and the kind of philanthropic attention that manifests as naming opportunities and endowed chairs. The college or university that can credibly say "our infrastructure is ready for the work you want to do" is both winning a recruitment conversation and positioning itself for a decade of research momentum that will be difficult for underprepared institutions to replicate.

Strengthening the Workforce Pipeline

A second infrastructure story is unfolding alongside the research computing conversation, and in states like Wisconsin, where I work, that story may ultimately matter more than the faculty recruitment one. The national data center buildout—driven by AI, cloud computing, and the digitization of everything from health care to agriculture—is generating an enormous demand for people who know how to build, operate, and manage large-scale computing infrastructure.^Footnote2

Demand extends beyond software engineers to include data center technicians, power and cooling specialists, network engineers, systems administrators, and the operations professionals who keep these facilities running around the clock. These are good jobs: well-compensated, locally anchored, and not easily exported. And right now, the pipeline producing them is nowhere near adequate to meet the demand that is already materializing.

Wisconsin is reasonably well-positioned for data center investment; it has a favorable climate, available land, and energy infrastructure with room to grow. But physical advantages only go so far. The regions that win the long game in this buildout will be the ones that can also answer the workforce question. And answering that question runs directly through universities and technical colleges. Ultimately, the answer will be determined by the willingness of higher education institutions to treat data center operations as a discipline worthy of serious curricular investment.

The opportunity is clear: A college or university that builds meaningful AI infrastructure capability and pairs it with programs that train the next generation of people to operate that infrastructure is not just serving its own research mission. It is serving its region's economic development agenda, its local business community, and the students who will build careers at the intersection of technology and operations for the next thirty years.

Leading the Infrastructure Transition

I have been in enough infrastructure planning conversations to know that the instinct, when faced with a capital-intensive challenge, is to wait for certainty—to see which approaches gain broader adoption, to watch what peer institutions do, and to find a moment when the path forward is obvious enough to justify the investment. That instinct is understandable. It is also, at this particular moment, a strategy for arriving late to a competition that has already been decided.

The institutions that will be positioned to recruit top AI researchers in three years are making infrastructure decisions today. Not reckless ones, but thoughtful, phased, and strategically sequenced investments in power capacity, cooling modernization, high-density compute, and the network architecture to support it. They are treating their data centers not as utilities to be maintained but as competitive assets to be developed.

For higher education technology leaders, the practical agenda is clear: audit your current infrastructure honestly, understand the gap between what you have and what serious AI research workloads require, build the internal case for investment with language that connects physical infrastructure to faculty recruitment, student outcomes, and research revenue, and move before the window narrows further.

The best AI researcher in the world is choosing between institutions right now. Whether yours is seriously considered in that conversation depends on decisions that are being made, or deferred, in the months ahead.

The pause before answering the infrastructure question is information. The question is what you do with it.

Notes

1. Peng Shu et al., "Survey of HPC in US Research Institutions," arXiv, June 24, 2025. Jump back to footnote 1 in the text.↩
2. Douglas Donnellan et al., Uptime Institute Global Data Center Survey 2024, (Uptime Institute, July 2024). Jump back to footnote 2 in the text.↩

Steven Goodman is Senior Director of Technology at Marquette University.