AI Inference Costs Fall 60% in a Year, but Jevons Paradox Sends Corporate Bills Soaring

Anthropic says the cost of generative AI inference has fallen by about 60% over the past year, a decline that should have lowered barriers to enterprise adoption. But under the Jevons paradox, cheaper resources spur greater demand. Companies that continue to pay passively by the token may see their total bills rise rather than fall, making cost governance critical to deploying AI at scale.

Recent discussions show that enterprise AI spending has continued to climb despite the 60% annual drop in inference costs, mainly because call volumes, context lengths and automated workloads are all increasing. Experts recommend matching large, open-source or small models to the complexity of each task and actively managing token usage. Current reports have not disclosed specific billing amounts or the date of the event.