Wuhan Court Makes AI Token Costs Part of Copyright Damages Math

A court in Wuhan has become the first in China to fold the cost of running AI tools into a copyright damages calculation, treating token usage and licensing fees as factors in what an infringer owes. The ruling, reported by The Decoder, marks the first time production costs tied to generative AI have entered the damages math in a Chinese infringement case. It effectively puts a price tag on the machine side of content creation, not just the human work behind it.

The mechanics matter more than the headline number. Tokens are the units that AI models consume when processing prompts and generating output, and platforms bill for them; licensing fees cover access to the underlying tools. By admitting both into the damages ledger, the Wuhan court is treating AI-assisted production as a real, quantifiable input cost rather than an invisible one. That is a notable shift for a legal system that has mostly measured copyright damages against the value of the copied work itself.

The decision lands inside a broader Chinese effort to write rules for AI-generated and AI-assisted works. Beijing has been moving on several fronts — labeling requirements for synthetic content, draft rules on training data, and a steady trickle of court decisions testing how existing copyright law applies to machines. Wuhan's contribution is narrower but concrete: it gives litigants a framework for arguing that the expense of generating a work is part of what was taken.

That framework cuts both ways. Rights holders can point to production costs to argue for higher awards, since a defendant who copied AI-assisted output avoided both the human labor and the compute bill. Defendants, meanwhile, may use the same logic to argue that token fees and subscriptions are routine operating expenses, not a measure of creative value. How Chinese courts weigh those competing readings will shape settlement behavior long before any statute catches up.

It also raises a question U.S. and European courts have largely dodged: whose costs count when the work was made with a model trained on someone else's data? If a plaintiff's damages include fees paid to an AI vendor, and that vendor's training set is itself under litigation, the accounting gets tangled fast. Wuhan has not resolved that, but it has opened the door to treating the AI supply chain as part of the damages picture.

For now, the practical takeaway is that Chinese plaintiffs have a new line item to argue and defendants have a new one to contest. Watch whether later rulings in Wuhan or higher courts cite this case when setting damages in disputes over AI-assisted text, images or code — and whether platforms start keeping token and licensing records with litigation in mind.