AWS AI Revenue Scale

Amazon just put a number on something the cloud industry had mostly described in vibes: its Amazon Web Services artificial intelligence business is already running above a $15 billion annual pace in the first quarter of 2026. That was the first time Amazon disclosed a direct revenue figure for that work. (reuters.com) That number came from Chief Executive Officer Andy Jassy’s shareholder letter on April 9, 2026, and he paired it with another one: Amazon’s custom chips business, which includes Graviton and Trainium processors, is now above a $20 billion annual revenue run rate. The pairing matters because cloud artificial intelligence demand and the chips that power it now sit in the same capital cycle. (reuters.com) Amazon Web Services is not a side business inside Amazon. It booked $128.7 billion of revenue in 2025, so a $15 billion artificial intelligence run rate means this is no longer a lab project buried inside a giant cloud division. (reuters.com) A revenue run rate is not the same thing as booked annual sales. It means if the current quarter’s pace kept going for four quarters, the yearly total would be above $15 billion. (reuters.com) The reason cloud companies care about that threshold is simple: once customers are spending at that scale, they stop treating artificial intelligence like a demo and start treating it like payroll software. They want contracts, uptime targets, and predictable bills. (reuters.com) In cloud artificial intelligence, the expensive step is often inference, which is the moment a model generates an answer after a user sends a prompt. Amazon Web Services now sells different Bedrock service tiers for that step, including Flex for cheaper non-urgent jobs and Priority for faster mission-critical jobs. (aws.amazon.com) Amazon Web Services also prices Bedrock in ways that push customers to think like operations teams instead of experimenters. Its pricing page offers on-demand use, batch processing at 50% lower price than on-demand for some models, and reserved or provisioned capacity for steadier workloads. (aws.amazon.com) When thousands of companies share one cloud, tenant isolation becomes the rule that keeps one customer’s data from leaking into another customer’s job. Amazon’s own guidance calls isolation a foundational element of software-as-a-service systems, and its Nitro Enclaves product creates isolated compute environments for sensitive data like health and financial records. (docs.aws.amazon.com) (aws.amazon.com) Developer tooling becomes a money issue at this size too. Amazon said this week that Bedrock Projects lets teams attribute inference costs to specific workloads inside Cost Explorer, which is the kind of feature companies ask for when finance teams need to explain a surprise bill, not when engineers are just playing with a chatbot. (aws.amazon.com) Jassy said the artificial intelligence numbers are “ascending rapidly,” and Reuters reported Amazon already has customer commitments for a substantial portion of the chips business. That means the next fight in cloud artificial intelligence is less about who can launch a model demo first and more about who can deliver capacity, isolation, and cost control at industrial scale. (reuters.com)