In a major move in the AI industry, Meta has signed a deal to use millions of Amazon’s Graviton chips to support its growing AI systems.

Graviton chips are CPUs (central processing units), not GPUs. While GPUs are still mainly used to train large AI models, CPUs are becoming more important after training — when AI systems are actually used.

This is where things are changing. Modern AI tools and agents now perform tasks like real-time thinking, writing code, searching, and handling multi-step operations. These tasks require strong and efficient computing, which CPUs like Graviton are designed to handle.

With this deal, Meta is sending more of its business back to Amazon Web Services (AWS), instead of relying on competitors like Google Cloud or Microsoft Azure. Previously, Meta had signed a $10 billion deal with Google Cloud, but AWS has now regained a strong position.

Amazon is also working on its own AI GPU called Trainium, which is used for both training and running AI models. However, many of these chips are already committed to another company, Anthropic, which recently signed a massive long-term deal with AWS.

This new agreement with Meta shows Amazon’s confidence in its own chip technology. The company is trying to compete with big players like Nvidia and Intel by offering better performance at lower costs.

Overall, this deal highlights a growing shift in AI — where not only GPUs, but also powerful CPUs are becoming essential for running advanced AI systems in real-world use.