Nvidia CEO Jensen Huang is signaling a significant shift in strategy, preparing investors for a renewed battle with Intel and AMD in the CPU market.
While Nvidia built its success on specialized GPUs for AI, Huang now professes a strong interest in generalist CPUs. He notes a dramatic ratio flip, with CPUs now handling a larger portion of computing tasks, especially as AI companies move from model training to deployment.
"We love CPUs as well as GPUs," Huang stated, assuring analysts that Nvidia's own data center CPU offerings, launched in 2023, are poised to outcompete rivals. He anticipates Nvidia becoming a major CPU maker.
CPUs are generalist chips handling diverse tasks, while GPUs excel at parallel processing of simpler calculations, crucial for AI data matrix operations. As AI "agents" perform complex tasks like coding and research, these operations increasingly rely on CPUs.
Nvidia's flagship AI server, the NVL72, currently features a 1:2 ratio of Nvidia CPUs to GPUs. Analysts suggest this ratio could shift, with CPUs potentially dominating or even replacing GPUs for certain agentic workloads.
Underscoring its CPU ambitions, Nvidia secured a deal with Meta Platforms for its Grace and Vera CPU chips. While Meta continues buying AMD CPUs, Nvidia aims to prove its CPUs offer superior data processing capabilities for AI-driven problems.
Nvidia is positioning the CPU not as the default foundation of compute infrastructure, but as a flexible architectural option. Further disclosures on Nvidia's CPU advancements are expected at its upcoming developer conference.