AWS AI revenue disclosed

Amazon just put a price tag on a business investors had mostly been guessing at: its Amazon Web Services artificial intelligence products are now running at more than $15 billion a year, according to Andy Jassy’s shareholder letter published on April 9, 2026. That is the first time Amazon has broken out a direct revenue number for this part of Amazon Web Services. (aboutamazon.com) (reuters.com) Amazon also said it expects to spend about $200 billion on capital expenditures in 2026, with most of that aimed at artificial intelligence infrastructure. In plain terms, that means more data centers, more networking gear, and more chips to run and train artificial intelligence models. (aboutamazon.com) (reuters.com) Amazon Web Services is Amazon’s cloud-rental business: companies pay it for computing power the way a factory pays an electric utility for power from the grid. Artificial intelligence has turned that business into a hardware race, because every chatbot, coding assistant, and image model needs huge clusters of servers before it can answer users in real time. (amazon.com) (cnbc.com) That is why Amazon keeps talking about its own chips instead of only buying chips from Nvidia. Jassy said Graviton processors, Trainium artificial intelligence chips, and Nitro networking cards together now have an annual revenue run rate above $20 billion. (aboutamazon.com) (geekwire.com) Graviton is Amazon’s general-purpose server chip, the part that handles ordinary cloud workloads like databases and web apps. Trainium is the specialist chip for teaching large artificial intelligence models, which is the expensive step where a model digests giant piles of text, images, or code. (aws.amazon.com 1) (aws.amazon.com 2) Nitro is the hardware-and-software layer that offloads networking, storage, and security jobs from the main server, like moving traffic control out of the driver’s seat so the engine can focus on speed. Amazon says that system is now used across most of its Amazon Web Services compute instances, which is why a chip business that sounds niche can show up as a multibillion-dollar line inside the cloud unit. (aws.amazon.com) (aboutamazon.com) Jassy’s argument is that these chips do two jobs at once: they lower Amazon’s own costs and they become products customers rent. He wrote that demand is still ahead of available capacity, and outside coverage of the letter noted that some customers were asking for as much Graviton capacity as Amazon could provide in 2026. (aboutamazon.com) (geekwire.com) The timing matters because investors have spent the past year asking whether the giant cloud companies were overspending on artificial intelligence. Amazon’s free cash flow fell from $38 billion to $11 billion in 2025, and Jassy tied that drop to a $50.7 billion increase in capital spending, primarily for artificial intelligence infrastructure. (geekwire.com) (aboutamazon.com) Amazon is not alone in that spending race, but this letter showed something new: the infrastructure is no longer being sold as a future promise with no numbers attached. Amazon now has one disclosed artificial intelligence revenue stream above $15 billion, one disclosed custom-chip run rate above $20 billion, and a 2026 capital plan of about $200 billion built around scaling both. (aboutamazon.com) (reuters.com)