Amazon’s AI numbers and bet

Amazon just put a price tag on its artificial intelligence push, and the numbers are bigger than many investors expected: more than $15 billion in annualized artificial intelligence revenue inside Amazon Web Services, and more than $20 billion in annualized revenue from its chip business. (reuters.com) Those are run-rate numbers, which means Amazon is taking the current pace of sales and stretching it across a full year, like turning one month’s restaurant receipts into a 12-month estimate. Andy Jassy gave those figures in his April 9, 2026 shareholder letter to show that the company’s artificial intelligence build-out is already producing real revenue, not just demos and promises. (cnbc.com) The reason Amazon is talking this way now is that Wall Street has been staring at one much larger number: about $200 billion in capital spending planned for 2026. Amazon said in its February 5, 2026 fourth-quarter results that this spending will go across the company, with a big share tied to artificial intelligence, chips, robotics, and satellites. (aboutamazon.com, s2.q4cdn.com) Jassy’s argument is that Amazon is not building empty warehouses full of servers and hoping customers appear later. He wrote that Amazon has already received commitments for a substantial portion of that spending and expects to monetize most of it in 2027 and 2028, while citing OpenAI’s commitment of more than $100 billion as one example of how large customer demand has become. (cnbc.com) The bottleneck is not customer interest. The bottleneck is capacity: data centers, networking gear, and especially electricity, because training and running artificial intelligence models is closer to operating an industrial plant than adding a few extra web servers. (aboutamazon.com, cnbc.com) Amazon said Amazon Web Services added 3.9 gigawatts of new power capacity in 2025 and expects to double total power capacity by the end of 2027. Jassy also said demand is still outrunning supply, which means Amazon believes the risk right now is underbuilding, not overspending. (aboutamazon.com, aboutamazon.sg) The chip number matters because Amazon is trying to avoid renting all of its brains from Nvidia. Its chip business includes Graviton processors for general cloud computing, Trainium chips for artificial intelligence training, and Nitro hardware that offloads infrastructure tasks, and Jassy said that business is growing at triple-digit percentages year over year. (cnbc.com, reuters.com) That gives Amazon two ways to make money from the same boom. It can sell artificial intelligence services through Amazon Web Services, and it can lower its own costs or even eventually sell more of the underlying chips and systems that make those services possible. (reuters.com, cnbc.com) The backdrop is that Amazon Web Services is already enormous before this new wave fully lands. In the fourth quarter of 2025, Amazon Web Services sales rose 24% year over year to $35.6 billion, and for full-year 2025 the cloud unit generated $128.7 billion in sales and $45.6 billion in operating income. (s2.q4cdn.com) So Amazon’s message is simple even if the spending is not: the company sees artificial intelligence the way it once saw cloud computing itself, as infrastructure that looks expensive before it looks obvious. On April 9, 2026, Jassy told investors Amazon is spending at this scale because demand is already here, the power is scarce, and the company wants to own more of the stack before rivals do. (reuters.com, cnbc.com)