For a decade, data-centre processors chased one metric above all others: more cores. Nvidia's Vera CPU team says the age of agentic AI is reversing that logic, forcing designers back toward single-threaded performance as AI agents run long, sequential reasoning loops that cannot simply be spread across dozens of cores.
Speaking to Wccftech, the Vera team described its next-generation data-centre CPU as a conductor for AI workloads — orchestrating the long reasoning chains that autonomous agents now perform, where the speed of each individual step matters more than raw parallel throughput. That is a sharp break from the chiplet strategies optimised for the cloud workloads of the last ten years.
The shift matters beyond one product. If agent-style workloads keep growing, the balance of power inside AI servers changes: the CPU that feeds and supervises accelerators becomes a performance bottleneck again, and chipmakers that spent years deprioritising single-core speed will have to compete on it once more.
Nvidia is positioning Vera at the centre of that transition, pitching the processor as purpose-built for the agent era rather than adapted from general-purpose server designs. Rival chipmakers are making their own versions of the same argument, and data-centre buyers will be watching benchmark results as the first Vera systems ship.
