# By 2045 US higher education bifurcates rather than uniformly shrinks: the top ~100 institutions grow, the middle hollows into teaching-only, and the bottom third of four-year schools merge or close — with regional decline tracking the demographic map more than AI itself (assessment as of 2025)

Demography, not AI, is the load-bearing wall. US high-school graduates peak around 3.9M near 2025–26, then the post-2008 birth-dearth cohorts hit college age from roughly 2029–2037. That decline is regionally lopsided: the Northeast, Midwest, and Plains lose on the order of 10–20% of college-age population while Texas, Florida, Tennessee, Georgia, and the Carolinas keep growing. So "grow or shrink" is already answered regionally before AI adds anything.

AI acts as an accelerant and a sorter, not the primary cause. It unbundles the campus-credential product first at the mid-tier: AI tutoring plus skills-based hiring erode the value proposition of non-selective private and regional public colleges whose degrees no longer gate employment — precisely the schools in demographically shrinking states whose legislatures won't backfill the losses. Meanwhile research flagships, land-grants, and Sun Belt/SEC growth-tier institutions gain pricing power from population inflow, athletics-attention economies, and their pivot toward AI research hubs and workforce pipelines. The middle loses its research identity first and becomes teaching-only — the most AI-exposed tier.

The surviving university of ~2045 is fewer in number, bigger, and more regionalized: a research engine with an experience-economy shell (residential life, stadium, brand) and credentialing demoted to a trust layer, with teaching split between a prestige faculty core and a model-supervised adjunct tier. Region matters, but it works through tier: the North's research flagships still hold while its teaching schools hollow first.

What would change this: if both selective privates and growth-tier flagships show flat or declining yields and international demand despite AI tools, the disruption runs deeper than bifurcation — the degree itself is devaluing, and even the growth tiers' tuition monopoly breaks under AI-native credentialing.