Amazon has secured a significant victory in the cloud chip wars, announcing that Meta has signed a deal to use millions of AWS Graviton processors to power its growing AI workloads. The timing of the announcement was pointed — Amazon dropped the news just as Google Cloud Next wrapped up, a move that read as a direct jab at its cloud rival. The deal is particularly notable because Meta had shifted significant business to Google Cloud last August with a six-year, $10 billion agreement, having previously been primarily an AWS customer. Bringing Meta's compute spending back to AWS is a meaningful signal that Amazon's homegrown chip strategy is delivering results.The Graviton is an ARM-based CPU rather than a GPU, and the distinction matters.

While GPUs remain the preferred hardware for training large AI models, the rise of AI agents is shifting demand toward a different type of compute. Agents handling real-time reasoning, code generation, search, and multi-step task coordination create workloads that CPUs like Graviton — designed specifically for AI-related compute — are well suited to handle efficiently. Amazon CEO Andy Jassy has made price-performance ratios a central theme of his chip strategy, and the Meta deal gives AWS a high-profile proof point that its homegrown silicon can win against competitors including Nvidia's ARM-based Vera CPU, which targets the same agentic AI workload category.