AWS: $15B AI run‑rate and huge capex

Amazon is telling investors that its artificial intelligence business inside Amazon Web Services is already running above $15 billion a year, which means the company is no longer talking about artificial intelligence as a future product but as a business line with the size of a large software company. At the same time, Amazon says it expects about $200 billion of capital spending in 2026, up from roughly $131 billion in 2025, and Andy Jassy says most of that money is going into data centers and artificial intelligence infrastructure. Amazon Web Services is the part of Amazon that rents computing power, storage, and software over the internet, so when companies build artificial intelligence tools there, they are usually paying for the digital equivalent of factory floor space and electricity. That matters because enterprise artificial intelligence is expensive in two layers at once: first you train large models on giant clusters of chips, and then you keep paying every time customers ask those models to generate text, images, or code. Jassy has been arguing for a year that lower costs will not shrink this market but expand it, the same way cheaper cloud computing led companies to run more workloads instead of fewer. Amazon is trying to own the whole stack around that bet: the data centers, the networking, the custom chips, and the software layer that lets businesses use models from Amazon, Anthropic, Meta, and others through one service called Amazon Bedrock. The $200 billion figure also covers projects outside cloud computing, and Jassy specifically grouped artificial intelligence with chips, robotics, and low Earth orbit satellites, which shows Amazon is treating computing capacity as shared plumbing across several businesses. Those satellites are now branded Amazon Leo, formerly Project Kuiper, and Amazon says the network is designed to use more than 3,000 satellites in low Earth orbit to deliver broadband internet to places that terrestrial networks do not reach well. Amazon said in late March that it had deployed more than 200 satellites and had more launches scheduled, so the satellite piece is no longer a paper plan sitting next to the artificial intelligence buildout. The scale is what makes this announcement land. CNBC reported in February that Amazon’s 2026 capital spending plan was above analysts’ $146.6 billion estimate and above Alphabet’s projected $175 billion to $185 billion range, putting Amazon at the front of the spending race among the biggest cloud companies. There is also a demand signal behind it: OpenAI said in February that it was expanding its existing Amazon Web Services agreement by $100 billion over eight years, which gives Amazon a giant customer to help fill all that new capacity. So the cleanest way to read Jassy’s message is this: Amazon thinks artificial intelligence will look less like a feature inside apps and more like a utility business, where the winners are the companies that can finance the most chips, the most power, and the most buildings before everyone else catches up.