Anthropic and Amazon have entered into a new agreement, providing up to 5 gigawatts of computing resources for training and deploying the Claude model.
Anthropic and Amazon have signed a large-scale agreement aimed at deepening the partnership initiated in 2023. The main goal of the collaboration is to provide up to five gigawatts of new computing power, dedicated to training and deploying the Claude artificial intelligence system. As part of the alliance, Anthropic commits to investing over one hundred billion dollars over ten years in Amazon Web Services technologies. This decision definitively secures the cloud platform's status as the primary provider for the developer's critical workloads. The agreement covers the current deployment of core infrastructure and options for acquiring future generations of custom processors as they come to market, ensuring long-term technological synchronization of both corporations.
The financial commitments of the parties include large-scale direct capital investments. According to the terms of the deal, Amazon will immediately invest five billion dollars in the developer company, reserving the right to direct up to twenty billion dollars in additional investments in the future. This funding round builds upon previous tranches totaling eight billion dollars. According to Anthropic CEO Dario Amodei, corporate users are increasingly noting that the neural network is becoming an indispensable tool in their workflows. This trend dictates the need for accelerated infrastructure buildup to keep pace with demand.
Hardware-wise, the technological alliance relies entirely on Amazon's proprietary semiconductor developments, specifically on Graviton server processors and several generations of specialized accelerators from the Trainium family, up to the Trainium4 architecture. The companies have already successfully launched Project Rainier, which is officially recognized as one of the largest computing clusters in the world. It currently utilizes over one million Trainium2 chips. A significant volume of new capacities will be commissioned in the second quarter of the current year, and a large-scale launch of systems based on Trainium3 processors is expected later. The overall roadmap sets the task of bringing the total volume of launched Trainium2 and Trainium3 clusters to nearly one gigawatt by the end of 2026.
In addition to increasing raw computing power, the updated partnership involves closer software integration: the full version of the Claude Platform will become available to clients directly within the global AWS ecosystem. Organizations will be able to use the environment through their existing accounts, while retaining all familiar administrative controls and a unified billing system. The integration provides direct access to advanced features without the need to manage additional credentials or enter into external contracts, which significantly simplifies compliance with strict corporate governance requirements. An official announcement from April twenty-first specifies that the launch will take place very soon.
The need to implement such an aggressive strategy is dictated by Anthropic's unprecedented operational growth. Corporate demand for generative models accelerated sharply throughout 2026. The explosive surge in commercial adoption has led to the developer's projected annual revenue surpassing the historical mark of thirty billion dollars. This financial indicator represents more than a threefold increase compared to the revenue of approximately nine billion dollars recorded at the end of 2025. Currently, over one hundred thousand corporate clients are actively executing their workloads specifically through the Amazon Bedrock cloud service.
However, such high growth rates have given rise to serious operational challenges, creating critical load on the company's current architecture. Developers have noted a sharp surge in user activity among private consumers across all subscription levels: the basic free version, as well as premium Pro and Max tariffs. The enormous influx of traffic negatively affected the overall reliability and performance of computations for users of Free, Pro, Max, and corporate-level Team accounts, which was particularly noticeable during peak server load hours. To eliminate these critical bottlenecks, the new agreement aims to provide significant resources within the next three months, with the goal of achieving one gigawatt of total capacity by the end of the year.
Sources
Replies (0)
No replies in this topic yet.
