Summary
Bloomberg Tech sits down with Matt Gorman, CEO of AWS, for a wide-ranging interview on the state of Amazon’s AI business following the company’s strongest revenue growth in nearly five years. Gorman confirms that AWS’s AI segment has reached a $25 billion annual revenue run rate, with demand spread broadly across enterprise and startup customers rather than concentrated in a small number of frontier lab clients like Anthropic and OpenAI.
On infrastructure investment, Gorman says Amazon’s 2026 CapEx stands at $220 billion — up $20 billion — and will grow further in 2027. He attributes the higher spend partly to elevated memory and hardware costs and partly to demand that continues to outpace available supply. Crucially, he reveals that much of AWS’s compute capacity is already committed through end of 2027 and into 2028, with customers signing five-year contracts for guaranteed access.
The interview also covers the ongoing shift from training workloads to inference, which Gorman confirms is accelerating as models become more widely integrated into production applications. He addresses Amazon’s Trainium chip line, describing it as sold out through end of next year and positioning it as a cost-reduction tool for high-volume inference workloads — with AWS controlling the full stack from silicon to data center to inference serving. Gorman frames the combination of Trainium and Nvidia GPU offerings as complementary rather than competitive, giving customers flexibility to optimize cost and performance at the workload level.
📺 Source: Bloomberg Tech · Published August 03, 2026
🏷️ Format: Interview
