Amazon invests another $4 billion in Anthropic

Background

This was a further step in Amazon's relationship with Anthropic, the company that develops the Claude models, following Amazon's earlier investment in it.

What happened

On 22 November 2024 Anthropic announced an additional $4 billion investment from Amazon, which brought Amazon's total to $8 billion. Amazon remained a minority investor.^[1] AWS became Anthropic's primary training partner as well as its primary cloud provider.^[1][2]

Key details

Anthropic said it would work with AWS's Annapurna Labs on future Trainium accelerators, writing low-level kernels and contributing to the AWS Neuron software stack. A kernel here is a small program that runs directly on the chip. Neuron is the software stack that AWS provides for its own chips.^[1] Anthropic planned to train its most advanced foundation models on Trainium.^[1]

AWS said tens of thousands of customers were running Anthropic models on AWS. It named ADP, Intuit, Pfizer and Moody's among them.^[2] AWS customers also got early access, for a period, to fine-tuning new Claude models with their own data. Fine-tuning means further training an existing model on the customer's own data.^[2]

Training a frontier model needs a large amount of computing capacity, and the chips used for training are a major part of that cost and planning. A long-term arrangement that covers both chips and cloud, and not only investment, shapes how a model developer plans the training of its next models.

Fine-tuning, the process of further training a model on a customer's own data, is what the early access covered, so customers could adapt the new models to their own material before the feature was opened more widely.^[2]

Impact

The investment deepened a partnership that now covered both cloud infrastructure and model training. Anthropic's most advanced models were planned for Trainium, and AWS customers gained a route to adapt new Claude models to their own data.^[1][2]