Channel one is mechanical and instant: concentration. With Mag7 at ~32% of the index and the top ten near 35-38%, a 25% fall in AI names alone removes roughly 9% from the S&P before any second-order effects. Passive investors who 'own the whole market' discover they own a leveraged bet on seven balance sheets. There is no diversified hiding place inside large-cap US equity anymore.
Channel two is credit, and it moves slower but hits harder. The AI buildout's paper — SPV bonds, data-center leases, neocloud debt, vendor-financing vehicles — sits disproportionately inside insurance portfolios ($800B+ of life-insurer assets in illiquid private credit), pension funds, and money pretending to be yield. Marking that paper to market during a downturn forces seller behavior exactly when buyers disappear; the 2008 parallel is not the equity crash but the funding freeze. The SEC's exemption of these structures from post-2008 securitization rules means the transparency shock compounds the price shock.
Channel three is the macro feedback loop that makes this cycle different from dot-com. Because AI capex carried ~92% of recent GDP growth, falling capex directly produces recession conditions — layoffs in construction/power/chips, wealth-effect contraction among the top-10% who own 90% of stocks, and state/local tax shocks from data-center abatements gone sour (Ohio's forecast $136M became >$1.5B paid). The policy response — rate cuts, possibly explicit backstops — then determines whether the outcome is 2001 (contained, slow recovery) or 2008 (systemic, years-long).
Sequencing expectations, if the repricing comes: semis and AI-pure-plays first (Korea already showed the pattern — SK Hynix fell on its BEST-ever quarter); then neoclouds and Oracle-type intermediaries as refinancing windows close; then the SPV/credit complex as marks arrive; then broad indices via concentration math; finally the real economy via capex withdrawal. Historical benchmarks for magnitude: dot-com took NASDAQ −78% but left the S&P at −49%; a repeat scaled to today's earnings-backed core argues for S&P peak-to-trough in the −30–45% zone in the severe scenario, −15–25% in the contained-crash scenario, and sideways-with-rotation in the deflation base case.
The wildcards that could accelerate everything: the 10-year yield crossing 5% (multiple corpus sources name this THE tripwire); a failed lab IPO absorbing the year's risk appetite; Japan's bond situation forcing another Treasury-dumping episode; or simply one hyperscaler defecting on capex guidance — Cameron Dawson's prisoner's dilemma, where cutting spending is individually rational and collectively detonating.