Agentic AI Shifts the CPU-GPU Balance in Data Centers

Semiconductor Engineering reports that agentic AI workloads demand more CPU compute due to if-then-else branching logic, which GPUs handle poorly. The CPU-to-GPU ratio could shift from 1:4 to as low as 1:1, already driving up CPU demand and prices.

• Agentic workloads could shift the CPU-to-GPU ratio from 1:4 to 1:1
• Agents rely on if-then-else logic that CPUs handle better than GPUs
• MoE and reasoning models still depend heavily on GPUs
• Rising CPU demand is already pushing processor prices higher