The AI infrastructure buildout reminds me of the pandemic playbook — and Peloton is the cautionary tale.

Pandemic demand for home fitness exploded. Peloton responded: bought a manufacturer, committed $400M to build a factory in Ohio. Then the economy reopened. Everyone who'd ever buy a $2,000 bike already had one. Factory never finished. Stock went from $160 to single digits.

We're seeing similar dynamics now in AI capex. Hyperscalers are spending tens of billions on GPUs, data centers, and infrastructure. Valuations reflect perfect execution and sustained demand. But what happens when the initial wave of AI adoption plateaus? When enterprises realize they overbought capacity?

I'm constructive on AI long-term — the technology is real, the productivity gains are real. But the pace of buildout today assumes linear growth forever. History says that's rarely how it plays out. Watch for signs of utilization rates, pricing power, and whether demand can actually absorb this supply.

Still bullish on quality names with diversified revenue streams and pricing discipline. But the pure-play infrastructure bets? That's where the hangover risk sits.