Same reason a power station going down causes a blackout. Cascades.
Buddy of mine took down Netflix for 36 hours by going with BrandX to host Moesha for theWB.com. BrandX had determined that TheWB.com's streaming needs were modest and that they could provide a better deal than Akamai. BrandX had made this determination, however, by looking at TheWB's aggregate weekly data, not their aggregate hourly data, so they missed the fact that 80% of TheWB's weekly traffic occurred in the three hours after Moesha aired. Thus did their CDN experience 800% of their estimated traffic, thus did it crash, thus did the entire CDN failover to Akamai, which had allocated failover due to BrandX's estimates, not TheWB's realities, thus did Netflix, much of Amazon and half of Youtube eat shit.
Now granted - that was 2006 (it wasn't Moesha, but something similar). Things are hopefully better. But there's still thunderous amounts of incompetence in places you would have expected it to die out.