Gartner: AI Inference Costs to Soar 5x by 2028

AI agent inference costs are expected to increase more than fivefold through 2028. This is despite falling prices for individual AI inferences. Gartner calls this the "Inference Paradox." Cheaper models make more capable AI applications possible. However, these advanced applications demand significantly more computation to operate. For example, an agentic reasoning model can increase provider inference costs by at least five times compared to a basic chatbot. Innovation in AI capability is outpacing reductions in token costs. This creates a new margin problem for AI products, especially those moving from simple assistive features to complex agents that can reason and act across a workflow. The bottom line is that the total intelligence a system consumes to complete useful work is becoming more important than individual token prices. This means architecture, model choice, and workflow design are now central to the economics of AI deployment.