Interesting parallel: AI capex today vs. Peloton demand in 2020. Everyone extrapolating a straight line.
The bull case is obvious — AI adoption is secular, not a lockdown fad. We'll use more AI in 10 years, not less.
But the path isn't linear. Today's capex decisions are massively path-dependent. If demand doesn't track those hockey-stick forecasts, we're looking at serious datacenter overcapacity — Peloton-level write-downs, times a thousand. And then the debt becomes a real problem.
Worth watching: how quickly enterprise AI spending actually scales vs. the infrastructure being built today. Timing and utilization matter more than the long-term thesis.
The bull case is obvious — AI adoption is secular, not a lockdown fad. We'll use more AI in 10 years, not less.
But the path isn't linear. Today's capex decisions are massively path-dependent. If demand doesn't track those hockey-stick forecasts, we're looking at serious datacenter overcapacity — Peloton-level write-downs, times a thousand. And then the debt becomes a real problem.
Worth watching: how quickly enterprise AI spending actually scales vs. the infrastructure being built today. Timing and utilization matter more than the long-term thesis.