Compensated Datasets from Mozilla Data Collective goes live in seven countries; data providers keep 100% of the licence fee, and AI labs pay Mozilla a 5% platform toll.
Mozilla Data Collective is opening a paid marketplace for AI training data, where the people who build the datasets name the licence price and the AI companies that train on them pay Mozilla a separate platform fee. The product, Compensated Datasets, launches on Wednesday across seven countries. The economic split is the structural detail: data providers keep 100% of the licence fee, and the AI lab downloading a dataset pays Mozilla a 5% platform fee on top.
Mozilla Data Collective is the first social enterprise incubated by the Mozilla Foundation, the non-profit behind Firefox. It grew out of Common Voice, Mozilla's long-running volunteer effort to record speech in dozens of languages for AI training, where contributors donated short voice clips. The paid product is the next step: instead of donating, contributors and dataset owners can list their work and name the price.
The launch matters because Common Voice was free, and AI labs used the speech data without paying the people who recorded it. A paid marketplace with a creator-set price is a deliberate break with that pattern, and Mozilla is explicit about the framing: founder and CEO E.M. Lewis-Jong called the move a departure from "extractive models" of AI data sourcing toward "human agency and fair value exchange."
The named launch partners are Pangeanic, TAUS, Karya, Spotlite, ContentX Labs, and YUX Design, with the announcement listing "and others." Each sets the licence fee for its own dataset. The AI lab pays that fee, then pays Mozilla a separate 5% on top for hosting, payment rails, and the verification work the platform does on data providers.
Rollout is narrow by design. The product is restricted to verified providers in the United Kingdom, France, Japan, the Netherlands, Singapore, Spain, and the United States, with more regions to follow. A Mozilla Data Collective community update in April previewed the payments feature as a near-term release, and Rest of World reported earlier this month that three community datasets had already been listed for paid commercial licensing as a precursor to Wednesday's general availability.
Three questions sit underneath the launch. What "verified provider" means in practice, and who audits it. Whether the seven-country initial list is broad enough to matter to a model lab weighing a multi-language training run that needs data from regions not yet on the platform. And whether a 5% platform toll funds real governance (consent review, takedown, audit) or just rent collection: compensation and consent are not the same problem, and the platform pays for one of them.
MDC officially launched as a beta in Barcelona on 6 November 2025 at the 15th Mozilla Festival, with multi-million-dollar seed backing. The Manila Times syndication of the wire is one of several mirrors of the original GlobeNewswire release. The watch item is whether any of the named launch partners publish the price they set for a single dataset. The first list price is the only signal that says whether "creators set the price" is a product feature or a press line.