The CapEx Cliff Question: When Does AI Infrastructure Spending Peak?
The CapEx Cliff is the single most important risk factor in AI infrastructure investing. Hyperscalers (Microsoft, Google, Amazon, Meta) are spending over $200 billion annually on AI infrastructure. This spending drives the entire semiconductor supply chain — from ASML's EUV machines to Vertiv's cooling systems. The question every AI infrastructure investor must answer: when does this spending plateau or decline? Historical parallels (2000 Telecom CapEx, 2018 Crypto Mining) show that CapEx cycles end suddenly, not gradually. Closelook uses Sentinel Tickers as an early warning system to detect the peak before it shows up in earnings.
The Bear Case: Historical Parallels
2000 Telecom CapEx: Telecoms spent massively on fiber optic infrastructure in 1998-2000. When revenue growth didn't materialize fast enough, CapEx was cut overnight. Companies like Nortel, Lucent, and JDS Uniphase lost 90%+ of their value. The supply chain (Corning, Ciena) collapsed alongside them.
2018 Crypto Mining: GPU demand for cryptocurrency mining drove NVIDIA's revenue to record highs. When crypto prices crashed, GPU demand evaporated in a single quarter. NVIDIA's stock dropped 55%.
The AI CapEx cycle shares structural similarities: massive spending driven by future revenue expectations, supply chains ramping capacity based on extrapolated demand, and a dependency on continued spending growth to sustain valuations.
The Bull Case: Why This Time Might Be Different
Unlike telecom fiber (which was built on speculative demand), AI infrastructure spending is driven by measurable inference demand. Every ChatGPT query, every Copilot suggestion, every AI-generated image requires compute. Inference revenue is growing and measurable — it's not speculative in the way 2000 fiber capacity was.
The counter-counter: just because demand is real doesn't mean it justifies $200B+ annual CapEx. The question is whether AI revenue growth at hyperscalers exceeds the infrastructure spend. If AI services generate $50B but cost $200B in CapEx, the math doesn't work indefinitely.
How Closelook Monitors the Cliff
Sentinel Tickers are the early warning system. ASML's order book decelerating signals CapEx pullback 6-9 months before it hits. Advantest's test equipment orders reflect whether new chip architectures are entering volume production. Micron's HBM pricing reveals whether demand is real or inventory-driven.
The Weekly Signal composite score captures CapEx-related risk through the Macro and Participation dimensions. A regime shift from Green to Yellow often correlates with early signs of CapEx deceleration.