AWS AI revenue spotlighted

Amazon just gave investors a rare hard number for artificial intelligence inside Amazon Web Services: the cloud unit’s artificial intelligence business is now running at more than $15 billion a year. Chief executive Andy Jassy put that figure in his April 9, 2026 shareholder letter, which is the first time Amazon has broken out a direct artificial intelligence revenue number like this. (aboutamazon.com) Amazon Web Services is the part of Amazon that rents computing power and software over the internet, and artificial intelligence is becoming one of the fastest-growing things customers rent there. Amazon said Amazon Web Services revenue reached $128.7 billion in 2025, so a $15 billion artificial intelligence run rate already makes artificial intelligence a meaningful slice of the cloud business. (aboutamazon.com) That $15 billion number is not a full-year booked total. It is an annualized run rate in the first quarter of 2026, which means Amazon is taking the current pace of sales and projecting what that pace would look like over 12 months if it held steady. (reuters.com) Amazon paired that revenue number with an even bigger spending number. Jassy said Amazon expects to spend about $200 billion on capital expenditures in 2026, with most of that money going into artificial intelligence infrastructure such as data centers, servers, chips, and networking gear. (aboutamazon.com) (cnbc.com) Capital expenditures are the giant upfront purchases that let a cloud company add more capacity before customers use it. Amazon already spent heavily before this new plan, with operating cash flow at $139.5 billion over the trailing 12 months ended December 31, 2025, while free cash flow was squeezed by infrastructure buildout. (aboutamazon.com) (cnbc.com) The clearest sign that this buildout is not just speculative is a single line in the letter about Trainium4. Jassy said Trainium4 capacity is already sold out, which means customers have effectively reserved all of the available supply of Amazon’s newest in-house training chip before Amazon can make enough of it. (aboutamazon.com) Trainium is Amazon’s custom chip for training large artificial intelligence models, which is the expensive step where a model digests huge amounts of data and learns patterns. Amazon is building these chips so it does not have to rely only on outside suppliers such as Nvidia for every artificial intelligence server it deploys. (aboutamazon.com) (reuters.com) Amazon said its broader chip business, including Graviton, Trainium, and Nitro, is now running at more than $20 billion a year. Graviton is Amazon’s general-purpose server processor, Trainium is its artificial intelligence chip, and Nitro is the hardware-and-software system that offloads networking and security work so cloud servers can spend more time on customer workloads. (aboutamazon.com) That chip figure matters because it shows Amazon is not only selling cloud access, but also shifting the economics of the cloud underneath it. If Amazon can move more workloads onto its own chips, it can lower its cost per unit of computing and keep more of the profit inside Amazon Web Services. (aboutamazon.com) (cnbc.com) The timing also lines up with faster growth at Amazon Web Services. In the fourth quarter of 2025, Amazon Web Services revenue rose 24% year over year to $35.6 billion, which was faster than the 20% growth Amazon Web Services posted for the full year 2025. (aboutamazon.com) Put together, Amazon is showing investors two things at once. The company says artificial intelligence demand is already producing real revenue today, and it is spending at a scale that suggests it expects the bigger payoff to arrive in 2027 and 2028 as more of that new capacity comes online and gets filled. (aboutamazon.com) (reuters.com)