Amazon boosts Anthropic investment to $8 billion

Amazon announced Friday that it is investing an additional $4 billion in Anthropic, the company behind the Claude assistant. The deal brings Amazon’s total financial commitment to $8 billion and cements one of the most significant partnerships in the race to develop large language models.

The agreement is about more than capital. Anthropic has chosen Amazon Web Services (AWS) as its primary cloud partner for training and running its future models. The company will use Amazon’s Trainium and Inferentia chips, which were designed specifically for artificial intelligence workloads.

Money, cloud and chips in one alliance

Amazon had already committed up to $4 billion to Anthropic in September 2023. The new investment doubles that figure, although Amazon will remain a minority shareholder and will not have a seat on the startup’s board.

The investment comes as developing frontier models requires exceptionally expensive infrastructure. Training a model involves processing enormous collections of text, code and images for weeks or months, with thousands of processors working in parallel. Then comes inference, the process of generating a response each time a user sends a prompt to Claude.

Until now, most of that infrastructure has been associated with Nvidia GPUs. AWS wants Anthropic to serve as a large-scale demonstration that its own accelerators can handle the workload. Trainium is geared toward training; Inferentia is designed to serve already-trained models at a lower cost per query.

For Amazon, selling access to servers is not enough: it needs customers that consume AI computing capacity on an ongoing basis. Anthropic is one of the few labs capable of doing that at a scale comparable to OpenAI or Google DeepMind.

Claude becomes a core part of AWS

Anthropic’s models are already available through Amazon Bedrock, AWS’s service that allows companies to use different generative models without building the entire infrastructure from scratch. Claude has become one of the platform’s most in-demand options, alongside Amazon’s own models and offerings from other developers.

The new deal deepens that relationship. AWS will be the primary environment for Claude’s future development, while Anthropic gives Amazon a flagship model to attract companies seeking text assistants, document analysis, code generation or internal-process automation.

The company, founded in 2021 by former OpenAI employees, also maintains a significant relationship with Google, which had committed up to $2 billion to the business. However, making AWS its primary cloud partner shifts the operational center of gravity toward Amazon, without necessarily implying absolute exclusivity.

Amazon’s answer to the Microsoft-OpenAI axis

The deal has an obvious competitive dimension. Microsoft has invested around $13 billion in OpenAI and integrated its models into Azure, Windows, Office and GitHub. Google is developing Gemini with its own teams and selling AI capacity through Google Cloud.

Amazon does not have an in-house general-purpose model lab with the public profile of OpenAI or Google DeepMind. Its strategy combines internal development, a platform open to third-party providers and financial partnerships with outside companies. Anthropic is the most important of those partners.

The distinction matters to customers. A company that contracts for Claude through Bedrock can keep its data, security controls and billing within AWS, without having to negotiate directly with each model creator. In return, the growing integration between Anthropic and Amazon means that choosing a model is becoming increasingly tied to a specific cloud platform.

The real test will be chip performance

The $4 billion investment strengthens Anthropic as its research costs rise rapidly. But Amazon’s strategic outcome will depend on something less visible than the investment figure: whether Trainium can deliver competitive performance and costs compared with Nvidia GPUs for training the next generation of Claude models.

If Anthropic manages to scale its models on AWS chips, Amazon will gain a commercial reference that will be extremely difficult to replicate. If development continues to depend heavily on rival hardware, the partnership will remain valuable for distributing Claude, but it will not solve Amazon’s more ambitious goal: turning its silicon into a genuine alternative for generative AI infrastructure.