Amazon doubling down on AI
Amazon is publicly repositioning itself to rebuild shopping and operations around AI, robotics and custom silicon rather than just bolting features onto old systems. (TechCrunch frames the company's annual letter as a blueprint for massive AI and infrastructure spending.) (techcrunch.com) The letter and reporting say the plan includes plans for large-scale robot and drone deployments and even the possibility of selling Amazon’s in‑house AI chips to other firms, signaling Amazon sees operational tech as a monetizable asset. (theregister.com)
Amazon just used its annual shareholder letter to say the company is not treating artificial intelligence like a chatbot add-on. Andy Jassy said Amazon is rebuilding core pieces of shopping, cloud computing, logistics, and hardware around it while defending roughly $200 billion in capital spending for 2026. (aboutamazon.com) (cnbc.com) The first clue is where the money is going. Jassy wrote that Amazon’s free cash flow fell from $38 billion to $11 billion last year after capital spending rose by $50.7 billion, largely because the company is pouring cash into artificial intelligence infrastructure. (aboutamazon.com) (geekwire.com) A cloud data center is basically a giant rented factory for computing power. Amazon Web Services, which is the cloud division that rents those factories to other companies, is where much of this bet lives. (aboutamazon.com) (reuters.com) Jassy said Amazon Web Services’ artificial intelligence business is already running at more than $15 billion a year in revenue. He also said Amazon has customer commitments covering a substantial portion of the 2026 spending and expects to monetize most of that capacity in 2027 and 2028. (reuters.com) (cnbc.com) The second clue is chips. Instead of buying every critical processor from Nvidia and Intel, Amazon has spent years designing its own silicon, the tiny slabs of circuitry that act like the engines inside servers. (techcrunch.com) (aboutamazon.com) Jassy said Amazon’s chips business, which includes Graviton, Trainium, and Nitro, is now running at more than $20 billion a year in revenue. He also said two large Amazon Web Services customers asked to buy all available Graviton capacity for 2026, which Amazon declined. (aboutamazon.com) (theregister.com) Trainium is Amazon’s in-house chip for training and running artificial intelligence models, which is the expensive part where companies feed huge amounts of data into software so it can learn patterns. Amazon says the newer Trainium2 systems offer 30% to 40% better price performance than its own graphics processor-based cloud instances, and first-generation Trainium cut training costs by up to 50% on some workloads. (aws.amazon.com) That cost gap is why the chip story matters. Bloomberg reported that Amazon is considering selling those chips to other companies directly, which would turn a tool built to lower Amazon’s own bills into a product line that competes more openly with outside chip vendors. (bloomberg.com) (theregister.com) The third clue is robots. Amazon said in November 2025 that it had deployed more than 1 million robots across its operations network, and those machines now sort, lift, carry, and move shelves inside fulfillment centers. (aboutamazon.com) Amazon is also pushing drones back into the picture. Its Prime Air service says it can deliver packages weighing up to five pounds in under 60 minutes, and Amazon says more than 60,000 items are available for one-hour drone delivery in select cities such as Phoenix. (aboutamazon.com) (aws.amazon.com) Put together, the letter reads less like a defense of one expensive year and more like a map of the company Amazon wants to be by the end of the decade: a retailer that runs on robots, a cloud provider that rents out artificial intelligence capacity, and a hardware maker that may sell the chips it first built for itself. (aboutamazon.com) (techcrunch.com)
