Fears of an AI Bubble Are Growing. Should SMH Investors Be Worried?

The key takeaway is not that semis are cheap or expensive in isolation, but that the AI capex cycle is still being funded from cash flow, which materially lowers the probability of a classic leverage-driven unwind. That shifts the risk from a sudden balance-sheet event to a slower sentiment reset: if hyperscaler spend decelerates, the first-order impact will likely be multiple compression in the most AI-beta names, while the broader semiconductor basket should hold up better because a meaningful slice of demand is still being pulled by non-AI end markets.

The second-order winner is the infrastructure layer with pricing power and diversified demand streams. NVDA and TSM remain the cleanest expressions of the AI buildout, but the article understates how much the market is already rewarding the “picks and shovels” narrative versus actual end-market visibility; that creates a setup where any capex pause could hit the group faster than fundamentals roll over. By contrast, names with gaming, auto, and industrial exposure should be more resilient in a slowdown because they have less binary dependency on a single budget cycle.

The contrarian point is that the consensus obsession with an AI bubble may be miscalibrated in timing rather than direction. A bubble can exist at the subsegment level even if the broad sector isn’t universally expensive, which means the trade is likely dispersion rather than outright sector collapse. Over the next 3-6 months, watch hyperscaler guidance and cloud backlog commentary; a downgrade there would matter more for semis than any headline debate about valuation multiples.