If Your Faculty Aren't Burning Tokens, You Have a Problem. So Do Your Students.

Jensen Huang said something last month that should be pinned above every dean’s door in America.

“If that $500,000 engineer did not consume at least $250,000 worth of tokens, I am going to be deeply alarmed.” He then compared a professional who doesn’t use AI tokens to a chip designer working with paper and pencil. Not someone who’s old-fashioned. Someone who is broken.

That’s a strong framing. It’s also the right one.

Huang isn’t making a point about tool adoption. He’s making a point about resource allocation. Tokens, in his view, are infrastructure spend, the same category as compute, electricity, or bandwidth. You don’t ask whether engineers “prefer” to use electricity. You measure whether they’re using the infrastructure you paid for to generate the output you’re paying them to produce. If they’re not, something has gone wrong.

Higher education should be uncomfortable with this framing. It mostly isn’t, which is itself the problem.

What it means for faculty

A professor who spends 40 hours writing a literature review that an AI could scaffold in 4 hours isn’t being rigorous. They’re being inefficient, and calling it scholarship. The question isn’t whether AI changes the nature of research (it does). The question is whether faculty are deploying the available infrastructure to work at a higher level.

Think about what’s actually possible: AI-assisted literature synthesis, course design tested against learning objectives at scale, personalized written feedback on student work without burning out the faculty member providing it. These aren’t futuristic scenarios. They’re available now. They require tokens.

If your institution hired a faculty member at $120,000 a year and that person is spending research time on tasks that a $20/month subscription could handle, the ROI math is not good. That’s not a knock on the faculty member. It’s a systems failure, a failure to equip, train, and expect.

Should universities start tracking token consumption for faculty? That’s a reasonable question to ask out loud, even if the answer is complicated. What gets measured gets managed. Right now, most institutions aren’t measuring anything.

What it means for staff

Admissions, advising, financial aid, student services: these are information-intensive, communication-heavy functions that run on human hours. They don’t have to.

An admissions office using AI to draft personalized outreach, flag yield risks, and process inquiries has a structural advantage over one that isn’t. An advising center that uses AI to surface at-risk students earlier, prepare advisors for appointments, and scale written follow-up is doing more with the same headcount. The institutions not doing this aren’t preserving something valuable. They’re just slower.

The staff version of Huang’s alarm isn’t about token counts. It’s about whether people in high-volume roles are spending their hours on judgment and relationships (things humans are still good at) or on tasks that should have been automated a year ago.

What it means for students

This is where it gets urgent.

The employers hiring your graduates are not debating whether to use AI. They already do. The question they’re starting to ask during interviews is not “are you comfortable with AI?” It’s closer to “show me how you work with it.” Token fluency, knowing how to prompt, iterate, verify, and integrate AI output into professional work, is becoming a baseline expectation, not a differentiator.

Universities, in many cases, are still arguing about whether ChatGPT is cheating.

That debate has its place. Academic integrity matters. But if the outcome of that debate is a generation of graduates who were discouraged from developing AI fluency because their professors were worried about homework shortcuts, we’ve protected the short-term grade and sacrificed the long-term career. That’s a bad trade.

The students who graduate knowing how to consume tokens effectively, knowing when to trust AI output and when to push back, knowing how to use these tools to do work that would otherwise be impossible, those students will have an advantage. Universities that teach this intentionally will produce more of them. Universities that treat AI primarily as an integrity threat will produce fewer.

The ROI calculation has flipped

That’s the core of what Huang is saying. For a long time, the risk was spending money on AI and not getting value. That calculation has inverted. The risk now is not spending, not deploying, not developing the institutional muscle to use these tools at scale.

Higher education tends to be the last major institution to absorb technological change. There are structural reasons for that: tenure, governance, accreditation, the slow churn of curriculum revision. Some of those reasons are legitimate. Most of them are not sufficient anymore.

Here’s the question worth sitting with: If someone audited your institution’s AI token consumption relative to its payroll, what would the ratio tell them about your readiness for the next five years?