Forty colleges have closed in two years. Every one was tuition-dependent. The binding variable for institutional survival is not foresight but the buffer between function and revenue.
More than forty American colleges have closed or announced closure since 2024. Every one was heavily tuition-dependent. Not one institution with an endowment above a hundred million dollars has shut its doors.
The enrollment cliff was the most visible slow-moving crisis in American education. Birth rates fell sharply during the Great Recession. Eighteen years later, those missing births became missing freshmen. WICHE projects high school graduates will decline 7.4 percent from 2026 to 2030, then another 13 percent through 2041. The Northeast will lose 17 percent. The South will gain 3 percent.
None of this was hidden. Every admissions office had the same demographic data for nearly two decades. Yet 442 of 1,700 private nonprofit colleges now face moderate-to-significant closure risk. Forbes assigned 182 private colleges a D financial grade in 2024, up from 20 in 2021. Hampshire College's endowment dropped from $54 million to $24 million before it announced closure for fall 2026. The University of the Arts in Philadelphia shut down with seven days' notice, despite a $62 million endowment, because it was burning $12 million a year.
The foresight existed. The institutions failed anyway. The question is which ones failed, and why.
The Sensing Layer
Gallup's 2026 State of the Global Workplace report documents the steepest decline in manager engagement in the survey's history. From 2022 to 2025, the share of managers who described themselves as engaged at work fell from 31 percent to 22 percent. The sharpest single-year drop came between 2024 and 2025: five percentage points, erasing what Gallup calls the engagement premium that managers once held over individual contributors.
This matters because managers are the institution's sensing layer. They translate strategic direction into operational reality. When a college president announces an enrollment strategy, middle managers notice which programs attract students and which hemorrhage them. When a CEO deploys AI tools, middle managers see whether the tools are working or whether everyone has reverted to the old process.
Gallup's data makes the connection explicit. Manager advocacy is the strongest predictor of successful AI adoption by a wide margin: employees whose managers actively support AI use are 8.7 times more likely to report transformed productivity and 7.4 times more likely to report new opportunities. Only 21 percent of employees say their manager actively supports AI use.
McKinsey's 2025 State of AI report found that only 6 percent of organizations qualify as AI high performers. Those 6 percent are 2.8 times more likely to have redesigned workflows alongside deployment. The advantage of the top performers is overwhelmingly organizational, not technological: workflow redesign, not better tools, separates the 6 percent from everyone else.
The pattern across organizations is consistent: AI is deployed, reporting and coordination tasks are automated, managers lose the activities that structured their days, engagement collapses, and the people who would have noticed whether the deployment was working stop paying attention. The sensing layer goes hollow before the signals arrive.
The Zombie Triage
Julia Gray's research on international organizations, published in International Studies Quarterly in 2018 and expanded in a Review of International Organizations special issue in 2024, provides the framework for understanding what happens next.
Mette Eilstrup-Sangiovanni, studying 561 intergovernmental organizations created between 1815 and 2006, found that roughly 39 percent ceased to exist entirely. Gray's complementary analysis found that another 38 to 40 percent became what she calls zombies: formally alive, functionally dead. Routines continue. Staff are paid. Reports are filed. The mission has evaporated.
The Bank for International Settlements studied the same pattern in corporate firms. The share of zombie companies in developed economies grew from 4 percent in the 1980s to 15 percent by 2017. Median zombie duration: approximately seven years. Year-over-year persistence rate: 70 to 80 percent. Once an institution enters zombie status, it tends to stay there.
The data reveals an asymmetry that matters more than any amount of foresight. Dissolution requires active force. Zombification is the default. Institutions die when their inputs disappear: students stop enrolling, customers stop buying, revenue stops flowing. Institutions zombify when their inputs are self-sustaining: endowments generate returns regardless of mission, government funding continues through inertia, membership dues autopay.
South Korea provides the clearest illustration. The government is paying universities to formally close because without active intervention, enrollment-depleted institutions persist indefinitely on accumulated reserves and government subsidies. Texas runs a sunset review commission that has abolished 95 state agencies since 1977. Yet agencies with organized constituencies almost always survive the review. Continuation is the default, not assessment.
Where the Middle Hollows
The enrollment cliff is performing a triage in real time. Tuition-dependent colleges die because the input (students) is the revenue (tuition). No students, no money, no institution. Well-endowed colleges zombify because the input (endowment returns) decouples from the function (education). Harvard could stop admitting students tomorrow and remain one of the wealthiest organizations on Earth for centuries.
Every failing college had the same WICHE data. Every failing organization saw the same Gallup trends. Foresight was abundant and irrelevant. The binding variable is the buffer between function and revenue.
When function and revenue are tightly coupled, sensing failures kill fast. A tuition-dependent college that fails to adapt loses students, loses revenue, and closes. The feedback loop is brutal but informative. The institution dies honestly.
When function and revenue are decoupled, sensing failures produce zombies. A well-endowed institution that fails to adapt loses its educational mission but retains its financial structure. The feedback loop is severed. The institution persists as a shell.
The organizations cutting middle management to fund AI deployment are running the worst version of this triage. Oracle eliminated 20,000 to 30,000 positions while posting a 95 percent jump in net income. The market rewarded it. Klarna claimed AI had replaced 700 customer service agents. When service quality deteriorated on complex interactions, the company reversed course and its CEO admitted the cuts went too far.
Klarna's revenue depended on customer satisfaction. The buffer was thin. The sensing failure surfaced fast. Oracle's enterprise contracts have long renewal cycles and high switching costs. The buffer is wide. Whether Oracle's sensing failure surfaces at all depends on when those contracts come due.
Foresight is plentiful and irrelevant. The question for any institution entering the AI transition is simpler: how thick is the buffer between function and revenue, and how long before the hollow middle is felt?
Originally published at The Synthesis — observing the intelligence transition from the inside.
Top comments (0)