Nvidia CEO Jensen Huang: AI Computing Power Shift Threatens Chip Profit Margins
Dwarkesh's interview with Jensen Huang clearly outlines the pressures Nvidia is facing: the largest cost center for AI labs has shifted from researcher time to computing power capacity, directly increasing customers' motivation to seek alternative chip designs.
The core argument in the interview is that if customers were originally paying a high premium for an Nvidia chip, and competitors can offer "good enough" alternatives at a lower price, AI companies will gradually shift their workloads like fleet procurement. Huang emphasized that chips are not interchangeable commodities; Nvidia's performance, software stack, and total cost of ownership still have advantages.
Source: Public Information
ABAB AI Insight
The underlying issue in this debate is the shift in procurement logic in the AI industry from "buying research time" to "buying computing capacity." As computing power becomes the largest cost, buyers will seek to reduce chip expenditures just as large logistics companies push down vehicle costs, recalibrating the comparative advantages of custom chips, TPUs, Trainium, and various ASICs.
The real pressure is not just from increased competition, but from Nvidia's historically relied-upon software moat and high-margin structure beginning to face cost rationality of "good enough." Historically, many high-profit hardware companies have been gradually eroded from "best performance" to "lower TCO" by customers, and once large clients begin to scale their own research or shift to specialized chips, profit margins often come under pressure before revenue does.
This also explains why AI infrastructure is experiencing capital pressure similar to energy markets: when demand surges and supply is limited, buyers will prioritize finding cheaper bottleneck substitutes. Nvidia is currently facing not just competition from a single product, but a re-layering of the entire computing power procurement system.