Amazon’s 20‑Year AI Bet
Amazon’s two-decade investment in AI is converging across AWS and consumer products, turning long-running infrastructure work into a company-wide AI push that touches commerce, cloud, and consumer services. The narrative frames AWS as both an enabler and beneficiary of Amazon’s wider AI ambitions. (cnn.com)
Amazon’s cloud origin dates to the launch of Amazon S3 on March 14, 2006, which Amazon says established the scalable storage foundation it now uses to host large-scale AI workloads. (press.aboutamazon.com) Amazon published the item-to-item collaborative filtering paper in 2003 (authors Greg Linden, Brent Smith and Jeremy York), a production ML system Amazon cites as an early, at-scale machine‑learning deployment. (amazon.science) On Feb. 5, 2026 Amazon projected roughly $200 billion in capital expenditures for 2026 and told investors most of that spending will go to AWS data centers to support AI and related infrastructure. (cnbc.com) Nvidia disclosed a multi‑year supply arrangement to sell about 1 million GPUs and other networking gear to AWS with deliveries beginning in 2026 and extending through 2027, according to Reuters reporting. (money.usnews.com via Reuters) AWS and OpenAI announced a multi‑year, $38 billion agreement on Nov. 3, 2025, under which OpenAI gains access to EC2 UltraServers and “hundreds of thousands” of GPUs on AWS to run and scale its core AI workloads. (openai.com) On the consumer side, Amazon unveiled Alexa+ at its Feb. 26, 2025 Devices & Services event as a generative‑AI upgrade across Echo and other hardware, while AWS made Amazon Bedrock AgentCore generally available on Oct. 13, 2025. (aboutamazon.com) (aws.amazon.com) Amazon confirmed in a Jan. 28, 2026 internal update that roughly 16,000 corporate roles would be impacted as the company reallocates resources toward AI and efficiency improvements. (aboutamazon.com) AWS has been hiring senior AI leaders from rivals and pushing AgentCore as an enterprise, production‑grade agent platform (AgentCore SDK downloads reportedly exceeded 1 million), signaling a deliberate move to productize internal AI infrastructure for external customers. (webpronews.com) (constellationr.com)
