In an interview, NVIDIA's Vera CPU team admitted that in the past, cloud infrastructure prioritized core count at the expense of single-thread performance and IPC. With the rise of Agentic AI, this trend is now reversing. The team reviewed the evolution of data center CPUs: In the early 2010s, vendors like Intel focused on high clock speeds and fewer cores, delivering strong single-core performance but limited total throughput. As cloud service providers scaled up, processor vendors began emphasizing core count and adopted multi-chiplet designs, but at the cost of single-thread performance and IPC. Now, the rise of AI Agents requires CPUs to refocus on single-thread performance because the way AI Agents work demands frequent switching between model inference, code execution, and other tasks, many of which rely on sequential execution. As a result, single-thread speed, memory bandwidth, and access latency have become critical factors affecting AI system utilization. NVIDIA's Vera CPU is designed specifically for this shift, aiming to combine high single-thread performance, memory bandwidth, and overall throughput.