The Second-Order AI Bottleneck: Hyperscaler CapEx Is Shifting Pricing Power to Custom Networking, Switchgear & Baseload IPPs
Hyperscalers are amortizing AI server fleets over 5–6 years on their income statements even as accelerator architectures refresh every 12–18 months. If enterprise AI software monetization lags the impending $45B+ annual depreciation step-up, CFOs can defer non-committed data-center phase-2 buildouts within two quarters—turning "unbreakable" switchgear and cooling backlogs into double-ordering air pockets, while FERC cost-allocation rules slow co-located nuclear PPAs.