While all the focus so far has been on GPUs, the next bottleneck lies in CPUs.

CPU bottlenecks account for ~90% of execution time for the agentic AI workloads, as CPU is responsible for API calls, data cleaning, and managing context windows.

Before agentic AI, we only needed one CPU for every 12 GPUs. But the latest projections from Nvidia put that at 1 CPU for every 2 GPUs.

AMD is the market leader in data center CPUs, and the gap is expected to widen further with the launch of 2nm Venice chips. AMD should get a better share of the incremental revenue driven by agentic demand for CPUs.

While both Intel & Nvidia (Vera CPUs) are competing in this market, AMD is much better placed to win [full report linked below].