AWS CFO joked about the financial toll of AI. An internal recording reveals why

An internal staff recording captures AWS CFO John Felton joking that Amazon’s AI buildout has pushed the company into negative free cash flow, reflecting a dramatic reversal from recent years. Amazon is committing roughly $220 billion this year to data centers, servers, and power to support machine learning workloads, a scale that dwarfs past infrastructure bets. Leadership argues demand justifies the outlay - AWS revenue grew 37% in the quarter and the sales backlog exceeded $490 billion - while CEO Andy Jassy frames AI as a once-in-a-lifetime opportunity. Yet investors and analysts are wary as capital spending rises and expected returns are being pushed farther into the future.

Internal documents and staff interviews reveal why: complex timing and coordination problems across permitting, construction, power hookups, and hardware deliveries are leaving expensive GPU racks staged but unusable. Amazon has begun taking earlier ownership of racks - including multi‑billion‑dollar shipments of Nvidia GB300 gear - so cash is spent sooner while facilities await power and commissioning. Forecasts shift frequently, and what might be a routine supply buffer for a smaller firm translates into billions of idle dollars at Amazon’s scale. The tension pits long‑standing frugality and operational rigor against an unprecedented, cash‑intensive AI expansion whose payoff timing remains uncertain.