Capital · 14 Nov 2026
Who pays for the GPUs?
The argument that NVIDIA must finance the AI build-out needs a denominator. Compare supplier investments with the funds supplied by operators, customers and financing markets, and separate cash paid from future commitments.
Gartner put prospective AI-provider infrastructure spending across 2026–30 at roughly $7tn. That is a forecast for the sector, spanning multiple sources of capital. The owner of a site, the buyer of a rack, a project lender and an accelerator supplier are four different parties. Their exposure does not fit on one chipmaker’s balance sheet.
In three named operator financings, NVIDIA supplied about $5.5bn, roughly 0.8% of estimated global AI investment in the comparable nine months. That is a selected subtotal. Broader identified equity and convertible purchases were much larger, at least $84.5bn since early 2023. A lab can spend an investment on research, salaries or cloud bills. A funding argument needs to spell out precisely which cheques it is counting.