In another wild turn for AI chips, Meta signs deal for millions of Amazon AI CPUs

In a significant development within the tech landscape, Meta has entered into a partnership with Amazon, securing a supply of millions of AWS Graviton chips to meet its expanding AI requirements. This announcement, made by Amazon on Friday, highlights the growing reliance on Amazon's proprietary chips for advanced computing tasks. The AWS Graviton chips are ARM-based CPUs, designed for general computing applications, distinguishing them from GPUs, which are typically preferred for training large AI models. However, as AI agents begin to generate compute-heavy tasks including real-time reasoning and multi-step task management, the demand for such CPUs is rising. Amazon claims that the latest iteration of Graviton has been tailored specifically for AI workloads. This latest agreement marks a strategic shift for Meta, redirecting funds back to AWS rather than its competitors, such as Google Cloud. Notably, last August, Meta had committed to a six-year, $10 billion contract with Google Cloud, although its historical preference had been for AWS alongside some usage of Microsoft Azure. Interestingly, the timing of this announcement coincided with the conclusion of the Google Cloud Next conference, suggesting a competitive jab at Google as they also showcased their latest AI chip advancements. While Amazon has its own AI GPU known as Trainium, which is used for both training and inference, Anthropic recently secured a substantial deal to utilize these chips, committing $100 billion over ten years to run its operations on AWS. Ultimately, this collaboration not only demonstrates Amazon's capability to host major AI customers but also positions its chips as strong contenders against Nvidia's newly developed Vera CPU. Unlike Nvidia, which sells its technology to various enterprises and cloud services, AWS solely provides access to its chips via its cloud platform. In recent statements, Amazon CEO Andy Jassy emphasized the need for improved price-performance ratios in AI, signaling a competitive strategy aimed at securing more deals in this domain, while increasing pressure on their internal chip development team to deliver results.