In a surprising move that underscores the rapidly evolving demands of artificial intelligence, Meta has signed a deal to acquire millions of Amazon's homegrown Graviton processors, as announced by Amazon on Friday. These aren't the GPUs typically associated with AI training, but rather ARM-based CPUs designed for general computing tasks.
This deal signals a significant shift in the AI hardware landscape. While GPUs remain essential for training large AI models, the rise of AI agents—autonomous systems that reason, write code, and coordinate tasks—is creating a surge in compute-intensive workloads that are better suited to CPUs. Meta's choice to deploy Graviton chips for these agentic workloads suggests that the chip race is no longer just about GPU supremacy.
Amazon's Graviton series, originally built for cloud efficiency, is now being repurposed to handle the real-time reasoning and orchestration required by AI agents. This development highlights how AI's maturation is driving demand for diverse chip architectures beyond the traditional GPU-centric model.