AWS CEO Says AI Demand Huge, Capital Spending to Stay High

Matt Garman, CEO of AWS, discussed the company's growth drivers and AI-related capital spending in a recent interview with a16z. He said AWS revenue is around $160 billion to $170 billion, up 37% year over year, and that AI demand is huge, with capital spending not expected to slow in the near term.

Garman attributed growth to three layers: enterprise cloud migration, startup compute demand, and agentic workflows. He noted that cloud migration and AI demand are simultaneously driving AWS growth, and that large enterprises need full AI infrastructure tied to data, storage, security, permissions, and production systems. He estimated that about 30% to 40% of AWS revenue comes from companies that were once startups. AI-native startups may require massive compute from day one, so AWS reserves scarce GPU and AI accelerator capacity for them rather than selling all resources to large frontier labs. AWS plans to purchase 2 million Nvidia GPUs in the coming years.

Regarding agentic workflows, Garman said more teams want cloud environments designed for agents. AWS is optimizing agent calls across data lakes, databases, and services through offerings such as Agent Core, Bedrock, and AWS Context, which could create new demand for inference, storage, security, and permission management.

Custom chips are also key. Garman said Graviton delivers about 20% improvements in cost and performance, while Trainium now powers most Bedrock traffic, with its capacity largely sold out through the end of 2027. He also emphasized that AWS capacity is not highly concentrated in a few customers, with single-customer exposure typically in the single digits or lower.