The two moves describe the same underlying problem: the capital cost of stocking enough high-performance chips to meet AI demand has grown faster than the revenue those chips generate at current rental rates. A price increase attacks one side of that gap; a sale-and-leaseback attacks the other.
The sale-and-leaseback structure is worth unpacking. Amazon would sell physical Nvidia processors to a third-party financier and immediately lease them back, keeping operational control of the hardware while converting a capital expenditure into a recurring operating expense. The balance sheet shrinks; the cash-flow obligation does not. What changes is how the liability appears to investors and, potentially, how it is treated for accounting purposes.
The price increase for GPU rentals is the more direct signal. Hyperscalers have competed hard on compute pricing to win AI workloads; raising that price suggests either that demand has given Amazon enough pricing power to act, or that the financing burden has reached a point where margin can no longer absorb it. Possibly both.
What neither move resolves is the underlying rate of capital commitment the AI infrastructure build-out requires, and whether the revenue it eventually produces will justify the structure now being used to fund it.
Harvey Jr.