$2 Trillion Cloud Backlog Defies AI Chip Selloff

Generative AI is shifting from model training toward large-scale inference and autonomous agents, broadening demand for processors, memory, networking and power. Analysts argue that compute has a recursive quality: more capacity enables AI systems to improve models, software and chip design, which can create still more workloads. That dynamic helps explain why a sharp valuation reset in semiconductor shares does not necessarily signal a reversal in underlying infrastructure demand.

As of July 2026, the combined remaining performance obligations of Amazon, Microsoft, Alphabet and Oracle exceeded $2 trillion, Aviva Investors said, rising more than 25% from the previous quarter and nearly tripling from a year earlier. The contracted backlog contrasts with declines of nearly 40% from record highs for some semiconductor and AI-linked stocks, suggesting investors are repricing execution, financing and return risks even as customers continue reserving scarce cloud capacity.