A twenty-point jump in September hike odds reprices the AI trade the earnings just validated

Thursday paid for the numerator and Friday declined to charge for the denominator. Nvidia's own stock finished the week up 1% ^[8] against a Nasdaq up 0.9% ^[9], which is a tenth of a point of relative gain ^[18] on the most watched print of the quarter, while Salesforce took 22% in a single session ^[3]. Twenty-one points between one company's week and another's afternoon ^[19], which tells you the chip result was traded as evidence about software rather than as evidence about chips.

Jensen Huang's line that return on invested capital is now under a year ^[10] happens to be the cleanest available argument for why the policy rate matters less to a GPU than to a subscription: an asset that returns its cost inside twelve months hardly notices the rate you discount it at, and a software multiple resting on cash flows that arrive late in the decade notices a great deal. The sector's rate exposure does not sit where the capex sits.

The load-bearing detail, then, is Colette Kress saying that current-quarter growth will be driven primarily by non-hyperscaler customers, neoclouds such as CoreWeave and Nebius among them, plus enterprises ^[11]. That is a change in the identity of the marginal buyer, from firms that fund clusters out of operating cash to firms whose cost of capital is a quoted price, and the material does not tell us how those buyers finance a purchase, which is exactly the disclosure I would want before treating Friday's repricing as somebody else's problem. This is probably wrong, but I think a September hike reaches Nvidia's revenue line before it reaches Nvidia's margins.

Against that view: Warsh kept a hike on the table rather than promising one ^[5], 55% still leaves 45% of the distribution saying he does nothing ^[21], and stocks being little changed on Friday ^[7] may be the correct read of an event that alters one quarter's funding cost and nothing about Amazon's commitment to another 2 million GPUs ^[2], a pace near a million a year across the two years it covers ^[20]. Or rather, the more interesting version of the bear case is that the additive thesis is true and the buyers of the marginal cluster are the ones who pay for the rate, which means the good news and the bad news have the same address. CNBC's Investing Club, which holds the stock, raised its Nvidia target to $280 from $260 ^[15]; that is a price, and the value question is who funds the 2028 half of Amazon's order.

The thesis fails if a hike lands in September and Thursday's software gains hold anyway, in which case the discount rate was never the binding input and I mistook a funding story for a demand story.