Some AI investors trim chip-stock exposure, expecting slower hyperscaler spending growth

The key mechanism is not a chip-demand collapse but a change in who captures the margin pool. If hyperscaler capex growth slows even modestly, the market will likely pull forward the day when AI hardware revenue becomes less linear, and the highest-multiple semiconductor names will de-rate first because their current valuations imply sustained upgrade cycles and very high utilization. That is a classic duration trade: the immediate risk is multiple compression in NVDA/SMH/SOXX, while the fundamental pain may only show up over 1-2 quarters as order growth normalizes and gross margin narratives get more fragile.

The potential beneficiaries are the buyers of the boom: MSFT, AMZN, GOOGL, and META could see capex-to-revenue efficiency improve if they can slow spend without impairing product roadmaps. Second-order, this also helps cloud platform margins and free cash flow optics, which may support buybacks and a relative rerating versus semis. The flip side is that slower spend can signal weaker AI monetization, so hyperscalers are not immune; if they cut capex because usage growth disappoints, the entire AI complex can trade down together.

The contrarian view is that the market may be overestimating how fast hyperscalers can throttle spending. Existing commitments, data-center buildouts, and competitive pressure can keep orders elevated for several quarters even if management rhetoric turns cautious. The cleaner tell is not sentiment but guidance: if next round capex guides stay sticky while AI revenue remains opaque, the rally can re-accelerate; if capex growth rolls over and inventory build appears in the supply chain, the de-rating window extends 6-18 months. The trade should be sized as a relative-value expression, not an outright bearish call on AI adoption.