Data center CPUs see second wind from agentic AI

NEW DELHI: The artificial intelligence (AI) industry is seeing a shifting balance between training models and deployment for inference and applications, including agentic AI. While much of the discourse has been about the significance of graphics hardware (GPUs), central processing units (CPUs) are emerging to the fore again.

GPUs excel at parallel processing for massive matrix calculations, which can reduce training times from months to weeks or less. On the other hand, CPUs handle sequential tasks, data preparation, and orchestration, the kind of activities that demand low-latency and single-core compute.