Alibaba targets revenue from big Qwen AI users

Alibaba (BABA) plans to require major commercial users of the next version of its Qwen open-source AI model to share a portion of the revenue they generate from commercial deployments. The measure is expected to be imposed as soon as next week, based on plans described by people familiar with the matter. This would introduce a revenue-sharing layer for certain large-scale users while Qwen continues to be developed as an open-source model.

Until now, Alibaba’s (BABA) monetization has largely focused on customers that access its models via hosted APIs on Alibaba Cloud. Developers and enterprises have generally been able to download and run open-source model weights in their own data centers without paying additional licensing fees. The planned change would specifically target sizeable external deployments that commercialize the technology at scale.

The new policy would extend commercial licensing obligations to large external or self-hosted uses of the next Qwen model, adding a new revenue stream beyond cloud API billing. Smaller or non-commercial users are not described in detail in the reports, which focus on major users expected to fall under the revenue-sharing requirement. This creates a differentiated structure in which only substantial commercial deployments would be subject to the new terms.

Reports describe the move as a step toward monetising open-source AI models while retaining broad access to the technology. By asking only major commercial users to share revenue, Alibaba would preserve open availability of model weights for much of the ecosystem but seek direct participation in the returns generated by large-scale services built on Qwen.

Coverage frequently compares Alibaba’s planned approach with the licensing model used for Moonshot AI’s Kimi K3 model. Kimi K3 requires entities that sell the model as a service and generate more than $20 million in annual revenue to negotiate a commercial agreement. In these negotiations, Moonshot AI has sought revenue shares of up to about 30%.

The comparison highlights how open-source AI developers are experimenting with revenue-sharing frameworks to capture value from high-earning services. While Kimi K3’s thresholds and share levels are publicly described, Alibaba has not disclosed similar figures for Qwen. The reports note that Alibaba’s exact revenue-sharing rate and other detailed licensing terms remain unclear and are still under discussion.

The planned revenue-sharing requirement for major Qwen users is expected to be announced as soon as next week, but no formal policy text has been released. The final design could determine which categories of users qualify as “major” and how revenue will be calculated. For now, the main established elements are the intention to charge large commercial users of the next Qwen model and the timing indication for the forthcoming change.

With the new framework, Alibaba would join other AI developers that combine open-source releases with targeted commercial terms for high-scale deployments. The outcome of the ongoing discussions over percentages and conditions will shape how attractive the next Qwen model is for enterprises planning to build large services on top of it.