Gartner predicts AI inference costs per agentic workflow to rise over 5x by 2028

Gartner forecast that AI inference costs per agentic workflow will increase more than fivefold through 2028, even as foundational model token prices fall. The firm attributes this to the shift from simple chatbots to multi-step agentic systems that require continuous reasoning, tool use, and higher token volumes. Product leaders face an 'Inference Paradox' in which improving unit economics enable more complex, higher-cost applications without guaranteed proportional value.