Amazon will invest up to an additional $25 billion in human resources as demand for artificial intelligence grows

Amazon.com announced Expanded partnership With AI startup Anthropic late Monday, the company pledged up to $25 billion in new investments, including $5 billion immediately and up to another $20 billion tied to commercial milestones.

The deal adds to Amazon’s previous $8 billion investment in Anthropic, bringing its total potential stake to $33 billion. In return, Anthropic has committed to spending more than $100 billion on Amazon Web Services technologies through 2036.

The AI ​​company will have access to up to 5 gigawatts of computing capacity to train and deploy its Claude AI models on custom silicon from Amazon. Anthropic currently uses 1 million AWS Trainium2 chips, and will gain additional capacity for Trainium2 and Trainium3 as Amazon brings 1 gigawatt online by the end of 2026.

“Our dedicated AI processor delivers high performance at a much lower cost to customers, which is why it is in high demand,” Amazon CEO Andy Jassy said in a statement. “Anthropic’s commitment to running its large language models on AWS Trainium for the next decade reflects the progress we’ve made together in custom silicon, as we continue to deliver the technology and infrastructure our customers need to build with generative AI.”

Amazon’s latest AI partnership follows A contribution worth $50 billion to OpenAI’s $110 billion funding round two months ago, a deal that valued OpenAI at $730 billion upfront. The e-commerce giant expects to spend $200 billion on capital expenditures this year, with the majority of it allocated to developing artificial intelligence infrastructure.

Founded in 2021 by researchers and former OpenAI executives, Anthropic has grown its annual revenue to more than $30 billion. The company’s Claude AI models compete directly with OpenAI’s GPT series and Google’s Gemini.

Amazon’s custom chip roadmap includes Trainium3 processors released in December 2025 and the upcoming Trainium4, which AWS says will provide 2 exaflops of performance for processing FP4 data.