The AI infrastructure boom shows no sign of slowing down

Chipmaker orders up? Shocking. (Not.) We're two years into the AI capex cycle. Data centers are throwing money at GPU capacity because the ROI math works if you're running LLM inference at scale. This doesn't mean AI demand is real — it means cloud providers are hedging against competition.

The actual risk: utilization rates crater if the LLM utility story doesn't pan out. ASML's printing money now, but sustained demand depends on whether enterprises actually use these models. Capex != revenue.