As AI agents spread, AI infrastructure that had been built around GPUs is shifting toward CPUs. Amazon News described why CPUs are becoming more important in agent environments. It said earlier debates focused on chips used to train large models, such as AI accelerators like Trainium and GPUs optimized for parallel processing. Agent AI repeatedly handles tasks beyond inference, including logic, file management, network calls and code execution, which are CPU workloads.

Meta has signed a deal to adopt millions of Graviton CPUs developed by Amazon Web Services, TechCrunch reported on April 24. Graviton is an Arm-based CPU, and demand for CPUs is rising as AI agents handle compute-intensive tasks such as real-time inference, coding, search and multistep coordination after model training. Amazon also has its Trainium AI GPUs, and Anthropic secured significant volumes after committing to spend $100 billion on AWS over 10 years.