On Monday local time, Amazon (AMZN) announced a landmark agreement with artificial intelligence company Anthropic, immediately investing an additional $5 billion and planning to invest up to an additional $20 billion upon the achievement of commercial milestones. Amazon had previously invested $8 billion in the creator of the Claude large language model. In exchange, Anthropic has committed to purchasing over $100 billion in computing resources from AWS over the next decade and to deploying Amazon’s self-developed Trainium AI chips. Following the announcement, Amazon’s stock price rose nearly 3% in after-hours trading to close at approximately $250.
Under the terms of the agreement, Amazon’s additional $5 billion investment is an immediate “down payment,” with the subsequent up to $20 billion tied to whether Anthropic meets “specific commercial milestones.” Anthropic’s reciprocal commitment is even more striking: purchasing over $100 billion in computing resources from AWS over the next decade, covering the Trainium AI chip series and tens of millions of Graviton cores, and obtaining up to 5 gigawatts of computing capacity—equivalent to the energy consumption scale of approximately ten large data centers.
The direct driver behind the agreement is the severe computing power bottleneck Anthropic is currently facing. According to previous reports, the Claude service has recently experienced persistent outages, rate limiting, and performance degradation, leading some customers to switch to other AI platforms. Meanwhile, Anthropic’s annualized revenue has exceeded $30 billion, more than tripling from about $9 billion at the end of 2025 in just a few months. Its coding assistance tool, Claude Code, has attracted over 100,000 enterprise customers running Claude models on AWS’s Amazon Bedrock platform. Under the agreement, the new computing capacity will come online over the next three months, with total computing power deployment expected to reach nearly 1 gigawatt by the end of 2026.
For Amazon, this transaction carries significance far beyond equity investment. AWS CEO Andy Jassy emphasized that Anthropic’s commitment to running its large language models on AWS Trainium over the next decade confirms the joint progress the two parties have made in custom chip development. Trainium is Amazon’s self-developed AI training chip designed to rival Nvidia GPUs, and over 1 million Trainium2 chips have already been deployed at the Project Rainier data center in Indiana. Through this large-scale deployment with Anthropic as a “marquee customer,” Amazon has demonstrated to the market the technical feasibility and commercial reliability of its self-developed chips.
Furthermore, Anthropic’s committed $100 billion in AWS purchases is equivalent to nearly an entire year’s revenue for Amazon’s 2025 AWS annual revenue of approximately $100.8 billion, and the lock-in period of ten years provides AWS with exceptionally significant revenue visibility. Amazon is also actively developing its own Nova series of large models and, through the Bedrock platform, aggregating external top-tier models such as Anthropic Claude and Meta Llama, thereby constructing a dual-track AI strategy of “both self-developed and third-party.”
The multi-billion-dollar agreement between Amazon and Anthropic marks an escalation in AI industry competition—from “who has the better model” to “who has the more complete computing power–chip–ecosystem closed loop.” Cloud providers with self-developed chip capabilities and hyperscale data center resources are upgrading from passive computing power suppliers to active shapers of the industry landscape.
Among Amazon’s competitors, Microsoft has invested approximately $13 billion in OpenAI cumulatively since 2019, but the market’s long-assumed “exclusive priority for Azure on OpenAI computing power” is being eroded. Another major competitor, Google DeepMind, continues to lead in foundational research, but in terms of the speed of commercializing enterprise AI and the depth of ecosystem lock-in, it has yet to demonstrate a flywheel effect comparable to AWS’s “model supermarket” or Microsoft’s “Copilot office suite.”
As Anthropic’s annualized revenue exceeds $30 billion and the company plans to launch an IPO within the year, this agreement will serve as a touchstone for testing the deeply integrated “model company + cloud provider” model. Anthropic is currently valued at $380 billion, far lower than OpenAI’s $852 billion. However, secondary market data shows that demand for Anthropic’s stock has already surpassed that for OpenAI, with buyers for the first time willing to pay a valuation premium for Anthropic.
