Gartner forecasts fivefold rise in AI agent costs by 2028

by@VentureBeat

Summary

Gartner has projected that costs associated with AI agents will increase fivefold by 2028, despite a decline in token prices, due to what they are calling the Inference Paradox. This phenomenon occurs because, while foundational AI labs are making token generation more cost-efficient, they are simultaneously investing these savings into advanced models that demand much greater computational power, leading to higher costs per token. Additionally, as organizations increasingly adopt more complex tasks that utilize these AI agents, the overall token consumption rises significantly compared to basic chatbot interactions. To mitigate rising costs, Gartner suggests enterprises implement strategies such as automated spending controls and model routing to more economical options.

Analysis

Gartner: Gartner is a leading research and advisory company that delivers insights and forecasts on technology trends and their business implications. In this news, the firm has outlined the Inference Paradox, forecasting rising AI inference costs for agentic workflows through 2028 even as individual tokens become cheaper. It provides actionable recommendations for enterprises on cost control through observability, tiering, and value tracking. Will Sommer: Will Sommer is a senior director analyst at Gartner specializing in AI and emerging technologies. He is quoted in the report explaining how model labs reinvest efficiency gains into more expensive frontier models and how greater efficiency enables more complex, token-intensive agent workloads. Sommer advocates for practices like AI FinOps and automated model routing to manage enterprise spending. Cost Drivers: Foundational AI labs are channeling efficiency gains into next-generation models that require greater computational power and cost more per token. Workload Shift: Greater token efficiency is enabling more complex AI agent tasks that consume far more tokens overall than basic chatbot interactions. Enterprise Response: Organizations are unevenly adopting measures such as automatic spending controls, model routing to cheaper options, and stress-testing deployments against cost variability.

Categories

aiai_agentstechmachine_learningvirtuals
View Original Tweet