AI Demand Growth and Divergent Infrastructure Prices

Abstract

Strong AI usage coexists with differentiated service prices, capacity rents, and memory costs. We connect these signals through exact spending accounting and conditional capital recovery. Between September and October 1-30, 2026, public-platform tokens rise 32.44% and quote-valued spending rises 25.51% to 34.26%. Baseline spending growth remains strong but decelerates. Common-model mix contributes 14.67 percentage points, while the matched aggregate unit value rises 0.64% and the comparable-model Fisher quote index falls 2.62%. Model-level quote increases coexist with a negative aggregate weighted-quote contribution. A100 rents decline as H100 and B200 rents rise; this generation pattern holds across all six matched 28-day price windows. Fixed-product memory quotations rise 1.55% for DDR4 and fall 12.96% for 512Gb TLC NAND over the latest 28-day comparisons, with the same opposing directions at earlier October endpoints. Dated contract forecasts indicate positive but moderating aggregate increases and stronger enterprise SSD pricing. Across 18 eligible adjacent-month comparisons, platform volume and matched spending each rise in 17, while model-mix contributions vary by month and concentration. The evidence establishes strong platform-demand expansion alongside differentiated monetization, capacity pricing, and procurement conditions. The capital-recovery framework identifies how rental contribution and relevant input-cost exposure jointly determine the utilization required to cover fixed and capital burdens.

Author abstract from the research manuscript. Typography and line breaks have been normalized; the linked source may show a different manuscript version.