You Could Buy Nvidia for Its 106% Revenue Growth and 75% Gross Margin. But There's an Even Better Reason the AI Stock Has Room to Run.

Rubin shifts the profit pool from "best GPU" to "best rack architecture," which is a material change in who captures the AI capex dollar. That favors networking, power, and thermal-management names where the attach rate rises with every denser deployment; think ANET, VRT, and ETN as second-order beneficiaries over the next 3-9 months. By contrast, OEM/server assemblers such as SMCI and DELL face tougher economics if the platform vendor captures more of the bundle and pricing power migrates upstream.

The near-term setup is still positive for NVDA, but the bar is now high enough that the stock can underperform on anything short of perfect execution. The real risk is not demand; it is the physical rollout path: memory availability, datacenter power delivery, and shipment timing. If any of those bottlenecks slow, the market will compress the multiple even if revenue remains strong, because investors are paying for uninterrupted acceleration into the next 12-18 months.

The contrarian point is that the consensus is still using NVDA as the default AI proxy, when the better risk/reward may sit in the constraint layer. If the economics truly improve at the rack level, hyperscalers have an incentive to keep spending, but the incremental margin accrues to the picks-and-shovels around the racks rather than to every semiconductor name. That argues for relative-value exposure over outright chasing the flagship at an elevated multiple.