Amazon: AWS AI Infrastructure, Anthropic Stake and Timeline

A retailer that became a cloud company

Andy Jassy became Amazon's president and chief executive in July 2021, succeeding founder Jeff Bezos, who moved to executive chair. Jassy is not an outside hire: he built Amazon Web Services from the ground up in the 2000s, and AWS is the reason Amazon is relevant to AI history at all. Amazon reported $716.9 billion in revenue for 2025, most of it still retail, but AWS supplies roughly a third of the cloud infrastructure market, ahead of Microsoft Azure and Google Cloud. [2][1]

AWS's AI strategy runs on two tracks. Amazon Bedrock is the managed platform where customers rent access to models, including Anthropic's Claude and, since April 2026, OpenAI's models. Trainium and Inferentia are Amazon's own chips, built as an alternative to buying everything from NVIDIA. Amazon is, in other words, trying to be a supplier at every layer: the compute, the chip under the compute, and the storefront for the model running on it. [3]

The Anthropic position

Amazon's relationship with Anthropic has grown in four steps over three years. It began in September 2023, when Amazon committed up to $4 billion to Anthropic, with an initial $1.25 billion, AWS named Anthropic's primary cloud provider, and Trainium designated the training chip. A big cloud provider funding a lab to rent back its own compute was, by then, becoming the industry's standard shape. [4]

In November 2024 Amazon added another $4 billion, taking the total to $8 billion, and in April 2026 it added $25 billion more, five billion immediately at a $380 billion valuation and up to twenty more against commercial milestones. Amazon's total position reached $33 billion, and Anthropic committed over $100 billion of spending back to AWS across the following decade, with up to five gigawatts of capacity reserved. Anthropic's independence, on paper, sits alongside an increasingly specific dependence on one cloud provider's chips and cash. [5]

Supplying the industry, and fighting parts of it

Amazon's infrastructure role extends past Anthropic. When OpenAI's models reached AWS in April 2026 as Microsoft's cloud exclusivity ended, AWS became a distribution channel for a direct Bedrock competitor. In August 2026 AWS committed to 2 million more NVIDIA GPUs for 2027 and 2028, on top of a million already pledged, pushing its total commitment past 3 million chips, with 100,000 set aside for secure US government workloads. That scale of hardware demand has a price effect: AWS raised its GPU rental prices about 15 percent in a single weekend in January 2026, attributing it to supply and demand it expected that quarter. [6][7]

Amazon has also used the courts to defend the retail business AI agents are starting to route around. In March 2026 a federal judge temporarily blocked Perplexity's Comet browser from shopping inside password-protected Amazon accounts and ordered the data it had collected destroyed, an early court test of whether an agent acting on a shopper's behalf is legally the shopper or a separate party. [8]

One older Amazon business is a casualty of the same shift. In August 2026 Amazon announced the permanent closure of Mechanical Turk, the crowdsourced task platform Jeff Bezos once called "artificial artificial intelligence," alongside SageMaker Ground Truth and Augmented AI. Twenty-one years of human data-labeling infrastructure closed as dedicated AI-training labor markets absorbed the work instead. [9]

Before the money, there was a speaker

Amazon's AI history did not start with Anthropic. In November 2014 it put Alexa in a speaker called Echo, sold by invitation only and widely treated as a novelty at launch. It put an always-listening computer in ordinary living rooms years before any of them could hold a real conversation, and it normalized voice assistants as a product category well ahead of the large language models that would eventually make one worth talking to. [10]