# AI thins the US university map rather than emptying it — enrollment drifts ~19M → ~14–15M by the mid-2040s, institution count falls faster, and selectivity + licensure + region sort the survivors (Unverified: magnitudes)

The next two decades produce **bifurcation and thinning, not uniform collapse**. AI is an accelerant and sorter of forces already in motion — the demographic cliff and the erosion of the degree-as-signal — rather than the primary cause. Total enrollment drifts from ~19M toward ~14–15M by the mid-2040s, but the number of degree-granting *institutions* falls faster than the number of degrees conferred: campuses close and consolidate while output concentrates. The system thins at the median and grows at the top.

**Two-front squeeze.** Demographics: US births fell 4.32M (2007) → ~3.59M (2023) (CDC/NCHS); high-school graduates peak ~2025–26 and decline ~13% by 2041, with the Northeast, Midwest, and Plains hit first (WICSE/WICHE projections). Signal erosion: as AI absorbs routine cognitive work, the median bachelor's degree loses its labor-market premium first, while the elite degree's *sorting* function — allocation of people into scarce opportunities, decided largely at admission — becomes more valuable, not less. This makes the shift regressive: top programs tighten, median programs hollow out.

**Program axis.** Exposure tracks licensure and stakes, not subject. Nursing, engineering, and allied-health programs (state-board-gated, high-stakes) stay oversubscribed through ~2040; generalist BA tracks are the most exposed. An AI-verification layer — certifications, work samples, skills-first hiring — grows underneath the degree, most plausibly capturing the generalist lane. A counter-force holds the floor: the degree retains insurance value (graduates fare better in AI-driven layoffs) and legal