The AI Bubble Looks A Lot Like Dot-Com Bust

The important market mechanism is not that AI leaders are “expensive,” but that they have become the marginal driver of index performance. When a handful of mega-cap semis account for so much benchmark weight, passive and systematic flows create a reflexive loop: winners attract more inflows, and any disappointment triggers forced selling across QQQ/SMH, not just the individual name. That makes downside gap risk materially larger than day-to-day volatility suggests.

For NVDA, the base case remains strong fundamentals, but the stock is now trading like a long-duration asset: any hint that hyperscaler capex growth is normalizing can compress the multiple faster than revenue slows. SNDK is a more fragile expression of the same theme because storage is a higher-beta, more cyclical way to play AI demand; if inventory builds or order timing slips, it can underperform the broader AI complex even if compute spend stays resilient.

The next 1-3 months matter more than the next year: earnings, guide revisions, and cloud capex commentary are the real catalysts. The contrarian risk to the bubble thesis is that breadth stays narrow but earnings revisions keep rising, allowing concentration to persist longer than valuation models imply. The thesis is falsified if hyperscaler capex remains upwardly revised for another two quarters and semis keep beating without margin erosion; otherwise, concentration alone is enough to make any macro wobble hit these names first and hardest.