Nvidia’s print reinforces that AI infrastructure spend is still not rolling over, but the market should focus less on the quarter and more on who captures the economics of inference over the next 6-18 months. As custom ASICs migrate from pilot projects to production, hyperscalers and model labs can internalize more margin and reduce dependence on merchant silicon; that is a slow bleed to NVDA’s mix, not an abrupt demand shock. The near-term upside remains intact, but the long-duration multiple is vulnerable if investors start treating the business as a premium component supplier rather than the sole toll collector.

The bigger second-order winner is the platform layer: GOOGL, AMZN, and META can use in-house chips to lower unit compute costs and widen AI gross margin as token volumes rise. That matters because cheaper compute should accelerate workload adoption, but the incremental economics may accrue to cloud and app owners rather than the chip vendor. CRM is a relative beneficiary on the software side; if AI becomes a monetized feature rather than a capex arms race, enterprise software can rerate while infrastructure names are forced to defend spend intensity.

Macro is the tactical air pocket. A still-sticky inflation backdrop and Jackson Hole introduce duration risk for high-multiple tech, so even good AI fundamentals can be de-rated on a hawkish rates move over days to weeks. The thesis on NVDA is falsified if hyperscaler capex guides re-accelerate and no mix erosion shows up in the next 1-2 quarters; otherwise, the market may start pricing a slower terminal growth path by year-end.