Amazon has secured a significant agreement with Meta, leveraging its in-house developed chips. Meta has entered a deal to utilize millions of AWS Graviton chips to support its expanding AI demands, as announced by Amazon on Friday.
The AWS Graviton is an ARM-based CPU designed for general computing tasks, rather than a GPU.
Although GPUs are traditionally used for training large models, AI agents are changing the type of chip required for tasks such as real-time reasoning and code writing. AWS’s latest Graviton version is optimized for AI compute requirements.
This agreement redirects more of Meta’s financial resources back to AWS instead of competitors like Google Cloud. Previously, Meta struck a six-year, $10 billion deal with Google Cloud, although Meta mainly used AWS and Microsoft Azure before that.
AWS strategically timed the announcement of this deal to coincide with the conclusion of the Google Cloud Next conference, taking a virtual jab at its rival. Google also develops custom AI chips, releasing new versions at the event.
Amazon also produces its own AI GPU, the Trainium, used for both training and inference stages. However, Anthropic has recently signed a significant agreement to utilize many of these chips for future operations. Anthropic committed to a $100 billion expenditure over 10 years on AWS, focusing on Trainium use, with Amazon investing $5 billion in return.
The Meta agreement enables Amazon to highlight a major AI client as evidence of its CPUs’ capabilities. These chips compete with Nvidia’s new Vera CPU, also ARM-based and built for AI agentic workloads. The distinction is that Nvidia sells its chips directly, while AWS offers them only through its cloud services.
Amazon CEO Andy Jassy recently criticized Nvidia and Intel, emphasizing the demand for improved price-performance ratios for AI in his shareholder letter, signaling high expectations for Amazon’s chip development team.
This article may contain affiliate links.