AI in Higher Education Newsletter

July 21, 2026 · Vol. 25

A weekly brief for the Management Department, Isenberg School of Management, UMass Amherst. By Matthew D. Langenkamp / 雷邁德, prepared in collaboration with Thea 🪻✨.

← Back to archive · ← Back to Dictionary

AI Moves Onto Campus

This week’s AI story is infrastructure.

The higher-ed conversation has spent much of the past year on assignments, cheating, tutoring, and syllabus language. Those questions still matter. But a second layer is now visible: AI is moving onto and around campuses as physical infrastructure. Data centers, high-performance computing facilities, AI institutes, local models, institution-controlled systems, and vendor partnerships are becoming part of the university’s operating environment. The backlash is no longer only local: New York Gov. Kathy Hochul’s July 14 executive order pausing state environmental permits for new hyperscale data centers for up to one year shows that data-center politics are now state policy, not just campus politics.

That shift changes the question. We are no longer asking only, “Can students use ChatGPT?” We are also asking: where should AI run, who controls the system, who bears the energy and community costs, what happens to student data, and whether students experience AI as a tool, a tutor, a surveillance system, or a substitute for institutional attention.

Executive Summary

Three developments should be read together.

First, campus data-center controversies should be read alongside broader anti-data-center politics. New York’s one-year pause on new hyperscale data centers responds to the same worries showing up in campus protests: electricity demand, water use, local control, utility bills, environmental impact, and whether communities receive real benefits from projects built in their neighborhoods. The University of Michigan and Los Alamos National Laboratory’s proposed high-performance computing center in Ypsilanti Township, Oakland University’s proposed AI institute and data center, and Fisk University’s planned innovation center all show the same pattern: universities are being pulled into the physical politics of AI infrastructure.

Second, local and open-weight AI are making institution-controlled systems feel less theoretical. Rick Beato’s recent video makes the local-AI idea legible to a nontechnical audience, while Moonshot AI’s Kimi K3 release points toward a model ecosystem where more capability may be hosted outside the closed U.S. chatbot stack.

Third, student sentiment is not uniformly pro-AI. Students use AI, but many worry about dependence, authenticity, privacy, career disruption, and whether institutions are handling AI responsibly. For Isenberg, the answer is not to slow down. It is to design better: AI as professional tool, local or institution-controlled AI as governance option, and student skepticism as real evidence about trust.

1. The campus data-center debate has arrived

On July 14, Gov. Kathy Hochul signed an executive order creating what New York called the first statewide moratorium on new hyperscale data centers. The order pauses state environmental permits for up to one year while New York develops a regulatory framework around ratepayer protection, grid demand, water use, air quality, and community benefits. In plain English: the state is saying that AI infrastructure cannot be treated as a private real-estate deal if the public bears the energy, water, and infrastructure costs.

The University of Michigan / Los Alamos case is the clearest current campus example. Michigan Engineering describes a planned $1.25 billion high-performance computing and AI research facility with Los Alamos National Laboratory. One center would support Los Alamos work on national-security AI challenges; another would support U-M faculty, students, and partners working with Los Alamos on computational science, AI, materials, advanced manufacturing, and energy. Construction was expected to begin in 2026, with the facility fully operational in 2030.

That is the university-facing case for the project: scientific discovery, advanced computing, workforce development, energy, medicine, and national security.

The local politics look different. Recent local reporting from WXYZ describes strong opposition from Ypsilanti Township officials and residents, who object to the project’s scale, siting, community benefits, and the university’s exemption from local zoning control. AAUP’s spring 2026 Academe essay frames the same fight more sharply, connecting the project to militarization, labor politics, environmental harms, and the university’s investment priorities.

The Michigan protests and the New York moratorium are not identical. One is a campus-local dispute around a university / national-lab partnership; the other is a statewide permitting pause. But they rhyme. Both reflect a widening concern that AI’s physical footprint is being built faster than governance, community consent, and public-cost accounting can catch up.

The point for us is not to adjudicate every local claim from Amherst. The point is that AI infrastructure is no longer abstract. Universities that host or partner on large AI computing facilities inherit questions about electricity, water, land use, local consent, research purpose, national-security entanglement, and community benefit.

Teaching implication: AI literacy now includes infrastructure literacy. Students should understand that “using AI” is not just a software choice. It is also a question of compute, energy, governance, procurement, and institutional power.

2. Oakland and Fisk show this is a broader pattern

Michigan is not an isolated case.

Oakland University’s board voted in late June to move a proposed AI institute and data center into a year-long due-diligence phase. GovTech, summarizing local reporting, described a proposed 20- to 26-megawatt data center on 15 acres of campus land, with students and other attendees protesting during the board discussion. Oakland’s own February statement emphasized that the university had not made a final decision and was studying feasibility, partners, academic programming, site design, business planning, and infrastructure.

Fisk University offers a different version of the same story. Fisk’s $1 billion “Quantum Leap” master plan includes a 100,000-square-foot Innovation Center, with 30,000 square feet of academic space and 70,000 square feet of technology-center space on five undeveloped campus acres. Fisk presents the project as a responsible, community-centered framework for digital infrastructure and institutional sustainability. Local Nashville reporting shows community pushback, including petitions and concerns about whether a data center belongs on a historic HBCU campus.

The common thread is not “data centers are bad” or “universities should avoid AI infrastructure.” The better reading is that AI infrastructure changes the university’s stakeholder map. Students, neighbors, trustees, faculty, local officials, vendors, utilities, and communities all become part of the AI conversation.

For a business school, this is familiar territory. Strategy always becomes real in operations. The campus AI strategy is not only a statement of values; it is a capital project, a partnership model, a risk register, a stakeholder problem, and a governance design.

3. Local AI matters because it changes the governance options

Rick Beato is a musician, producer, and music educator whose YouTube channel reaches a large audience of musicians, teachers, and serious amateurs. His recent video, “I Was Right About AI”, is useful less as policy analysis than as a cultural signal. Beato argues by analogy to the recording studio: work that once required expensive centralized infrastructure eventually moved onto personal machines. He now sees AI moving in a similar direction, with more capable systems running locally or under direct user control.

The analogy is imperfect. Music production and frontier AI inference have very different capital requirements. A laptop recording studio did not require a 2.8-trillion-parameter model or a specialized inference stack. But the cultural signal matters: local AI is no longer just a hobbyist phrase from technical forums. It is entering the vocabulary of creators, teachers, students, and parents.

For universities, this becomes a FERPA and privacy issue. Student writing, grades, advising records, accommodations, and other education records should not casually move through general commercial AI systems. Local or institution-controlled AI can help protect sensitive work by keeping data inside governed systems, using university-approved retention rules, access controls, and audit practices.

That does not mean every AI use must be local. It means the institution needs options. Commercial cloud tools may be fine for some low-risk work. Sensitive educational records, student-support workflows, grading assistance, advising, and course-specific student data need a more careful posture.

4. Kimi K3 makes open-weight infrastructure less theoretical

Moonshot AI’s Kimi K3 is the week’s most important model-side signal. Moonshot describes Kimi K3 as a 2.8-trillion-parameter open-weight model with a 1-million-token context window, aimed at complex coding, knowledge work, visual tasks, and long-horizon agentic workflows. The company says full model weights will be released by July 27, 2026.

The careful word is provisional. Hosted access and benchmark claims are not the same thing as stable institutional deployment. Full weights, licensing details, independent testing, operational cost, safety behavior, and actual deployment requirements still need scrutiny.

Even with that caution, the direction matters. Powerful models are not going to remain confined to the closed U.S. chatbot stack. Open-weight and lower-cost models will keep pushing universities to ask which AI work should be hosted by vendors, which should be controlled by institutions, and which should be kept on local machines for privacy, experimentation, or pedagogical reasons.

This is where the campus data-center story and the local-AI story meet. At one end is extreme-scale infrastructure: Michigan / Los Alamos-style computing centers. At the other end is local or departmental AI: models running on personal workstations, lab machines, or institution-controlled servers. Universities will need to decide where along that spectrum different educational uses belong.

5. Student skepticism is part of the infrastructure problem

Student AI sentiment is more complicated than either panic or boosterism suggests. Inside Higher Ed’s Student Voice reporting found broad student use of generative AI for coursework, especially for brainstorming, tutoring-style questions, and studying. Lumina Foundation and Gallup similarly report that AI has become routine for many college students, even where campus rules are unclear.

But use is not endorsement. Students also worry about dependence, career effects, privacy, false accusations, and whether AI weakens authentic learning. Beato’s video adds an anecdotal but useful cultural signal: some young people do not merely dislike low-quality AI content; they dislike what AI seems to represent.

Isenberg probably has fewer outright anti-AI students than some humanities, arts, or media programs. Business students tend to be more pragmatic about tools that employers are likely to expect. But pragmatic acceptance is not the same as trust. A student can believe AI is professionally necessary and still object to being required to use tools that feel invasive, opaque, or corrosive of learning.

Design implication: do not build AI policy only for the enthusiastic user. Build for the student who wants career-relevant AI fluency but does not want every assignment, interaction, or learning space absorbed into AI infrastructure.

6. Attention is becoming the hidden benchmark

The Chronicle / Jagged Intelligence thread also surfaced a less comfortable point: some students in AI-teacher or AI-mediated learning models reportedly like the experience because their prior community-college experience felt anonymous or inattentive. The bot, at least, seemed to pay attention.

That comparison should make higher education uncomfortable. The threat is not that AI is better than good teachers. The threat is that AI can compare favorably with bad, overloaded, bureaucratic, or inattentive institutions. “Human teaching is irreplaceable” is true only when the human system actually notices the student.

For faculty, this is not an argument for replacing teaching with AI. It is an argument for defending the human side operationally: smaller feedback loops, visible instructor presence, timely response, meaningful office-hour access, and course designs where students are known by someone.

If AI systems are immediate, patient, and persistent, then human education has to compete on actual care, judgment, and relationship, not on slogans about human uniqueness.

7. Practical implications for Isenberg

For fall course design, the most useful move is to separate four roles for AI:

That frame is more useful than another generic syllabus paragraph. It tells students when AI helps, when it harms, when it is merely present in the system, and when the class is deliberately protecting human cognitive work.

The key sentence for the department may be this: AI is no longer only a classroom policy issue. It is becoming part of the university’s physical, legal, ethical, and relational infrastructure.

What to Watch Next Week

  1. Whether New York’s data-center moratorium becomes a template for other states, and whether campus disputes like Michigan / Los Alamos get pulled into that broader regulatory debate.
  2. Whether Oakland and Fisk move from feasibility and master-plan language into binding agreements, and how students and neighbors respond.
  3. Whether local or institution-controlled AI enters FERPA and privacy conversations more explicitly.
  4. Whether Kimi K3’s promised July 27 weight release produces credible independent testing.
  5. Whether fall syllabi distinguish AI as tool, tutor, infrastructure, and no-AI learning space.

Sources

← Back to archive · ← Back to Dictionary

Return to Dictionary All Entries (A–Z) For Students Other Writing