Amazon has secured a significant win with Meta, once again leveraging its own in-house chip technology. The company announced on Friday that Meta has agreed to use millions of AWS Graviton chips to support its expanding artificial intelligence needs. The Graviton is an ARM-based central processing unit (CPU), designed for general computing tasks, rather than a graphical processing unit (GPU). While GPUs remain the preferred option for training large AI models, the rise of AI agents built on these models is shifting demand toward CPUs. These agents generate compute-heavy workloads, including real-time reasoning, code writing, search functions, and managing multi-step tasks. AWS claims its latest Graviton version was specifically engineered to handle such AI-related computing demands.

This deal redirects a portion of Meta's spending back to AWS, away from competitors like Google Cloud. In August of last year, Meta signed a six-year, $10 billion agreement with Google Cloud, despite having historically been a primary AWS customer that also utilized Microsoft Azure. Notably, AWS timed the announcement just as the Google Cloud Next conference concluded, a move that seemed like a subtle jab at its cloud rival. Google, too, develops its own custom AI chips and unveiled new versions during the event.

Amazon also produces its own AI GPU, called Trainium, which is used for both training and inference—the stage where a model processes prompts after being trained. However, Anthropic recently secured a deal earlier this month that reserved many of those chips for years. The Claude maker committed to spending $100 billion over ten years to run its workloads on AWS, with a particular emphasis on Trainium, while Amazon agreed to invest an additional $5 billion into Anthropic, bringing its total investment to $13 billion.

Ultimately, the Meta agreement allows Amazon to showcase a major AI customer as a proof point for its custom CPUs. These chips compete with Nvidia's new Vera CPU, which is also ARM-based and designed for AI agent workloads. The key difference is that Nvidia sells its chips and AI systems to enterprises and cloud providers, including AWS, while AWS only offers access to its chips through its cloud service. Earlier this month, Amazon CEO Andy Jassy criticized Nvidia and Intel in his annual shareholder letter, arguing that enterprises demand better price-performance ratios for AI and that he intends to win deals based on that advantage. This places immense pressure on Amazon's internal chip development team, which we toured exclusively last month at their lab.