Amazon (AMZN) is buying billions in Nvidia (NVDA) AI systems it can't yet turn on, Business Insider reports

• AWS is expected to receive ~USD 5bln in Nvidia's GB300 server racks at US facilities before final deployment, according to August internal emails.

Deliveries of this kind, hardware arriving ahead of the power and cooling capacity needed to run it, have been a recurring feature of the AI buildout cycle, and the distinction that matters is between a supply bottleneck and a deployment bottleneck: the former constrains the chipmaker, the latter the hyperscaler, and the market has historically read them differently. For NVDA, racks shipped into staging count as recognized revenue at the point of delivery, so a buyer sitting on unpowered inventory flatters near-term shipments while raising the question of how much backlog is real end-demand versus capacity parked ahead of energization. For AMZN, the read runs through capex versus utilization: prior episodes where cloud operators absorbed hardware faster than facilities could be commissioned have eventually shown up as depreciation running ahead of AWS revenue, a gap the equity has historically punished when disclosed. The figures here come from internal correspondence cited by a single outlet, so confirmation sits with the next quarterly prints on both sides, specifically NVDA's shipments commentary and AMZN's capex and cloud capacity language. Worth noting that staged delivery before final deployment is also standard practice for large data center programs, so the telling detail is the scale relative to energization timelines rather than the existence of the staging itself. Follow-ons are any NVDA or AMZN comment on power procurement and data center commissioning schedules.