Alibaba May Introduce Revenue Sharing for Commercial Use of Next Qwen AI Model

Alibaba is reportedly preparing a new commercial licensing model for its next Qwen open-weight AI system, potentially requiring large companies to share revenue when they profit from offering the model as a service.

According to Reuters, citing two people familiar with the plans, Alibaba may require larger companies generating revenue from Qwen-based services to negotiate a separate commercial agreement with the Chinese technology giant.

The exact revenue-sharing percentage has not yet been finalized.

If implemented, the policy would mark a significant change from the licensing structure of Alibaba’s current Qwen3 open-weight models.

Alibaba currently releases Qwen3 models under the Apache 2.0 license, which generally allows commercial use, modification, and redistribution under its terms.

Developers using Qwen through Alibaba Cloud already pay for hosted access, but companies deploying the open-weight models on their own infrastructure have generally not been required to pay Alibaba licensing fees.

The next Qwen release could change that.

Open Source or Open Weight?

The development also highlights an important distinction in the rapidly evolving AI industry.

An open-weight model makes its trained parameters available for download, but that does not automatically mean the entire AI system is open-source or that every type of commercial use is unrestricted.

Under the Open Source Initiative’s Open Source AI Definition, users should be able to use, study, modify, and share an AI system for any purpose without requesting permission.

The definition also calls for access to information covering training data, relevant source code, and model parameters.

Chinese AI companies have increasingly released powerful models with downloadable weights.

Meanwhile, companies such as OpenAI, Anthropic, and Google continue to distribute most of their flagship commercial AI models through hosted and closed platforms.

Following the Kimi K3 Model

Alibaba’s reported strategy resembles the licensing system introduced by Chinese AI company Moonshot AI for its Kimi K3 open-weight model.

Kimi K3 allows users to download the model weights, but certain large commercial deployments are subject to additional conditions.

Under its published license, companies operating Model-as-a-Service businesses must negotiate a separate agreement with Moonshot once the combined revenue of the company and its affiliates exceeds $20 million during any consecutive 12-month period.

The rule also applies to derivative models based on Kimi K3.

Large consumer-facing deployments face another requirement.

Products exceeding either 100 million monthly active users or $20 million in monthly revenue must prominently display the Kimi K3 name, with exemptions for internal use and certain authorized services.

Sources familiar with Moonshot’s commercial arrangements told Reuters that these agreements may include revenue sharing.

One source said Moonshot could seek as much as 30 percent of revenue from certain partners.

Chinese IT services company Chinasoft International previously disclosed that it had entered into a revenue-sharing agreement with Moonshot, although the company did not reveal the percentage.

Cloud provider DigitalOcean also confirmed that it has a commercial agreement with Moonshot.

DigitalOcean CEO Paddy Srinivasan described the system as an open-source-style freemium model, where companies can begin using software at little or no cost but may eventually pay for large-scale commercial deployment, technical support, infrastructure services, or early access.

Open Models Are Not Free to Run

Downloading an AI model does not necessarily mean operating it will be cheap.

Large open-weight models still require massive computing infrastructure, especially when deployed at commercial scale.

Moonshot says Kimi K3 contains approximately 2.8 trillion total parameters, with around 104 billion parameters activated during inference.

Its mixture-of-experts architecture contains hundreds of specialized components, with only a portion activated for each token generated.

Alibaba is reportedly taking a similar approach with its Qwen3.8-Max architecture.

According to Reuters, the model contains approximately 2.4 trillion total parameters while activating roughly 95 billion parameters per request.

Mixture-of-experts systems can reduce computation by activating only selected parts of the model instead of the entire neural network.

Even so, operating models of this scale requires significant GPU resources.

Moonshot temporarily stopped accepting new Kimi K3 subscriptions in July after heavy demand placed pressure on its available GPU capacity.

For many businesses, self-hosting trillion-parameter models may therefore remain impractical.

This creates opportunities for cloud providers and AI infrastructure companies that specialize in inference, deployment, model optimization, and GPU hosting.

Dan Fu, vice president of kernels at Together AI, noted that value increasingly comes from making AI models more efficient and useful at the application layer.

Companies can differentiate themselves through better token efficiency, faster inference, optimized deployment, and specialized applications.

The Economics of AI Are Changing

Training frontier AI models is becoming increasingly expensive.

Research involving Epoch AI and Stanford researchers estimated that the cost of the most compute-intensive AI training runs has increased dramatically since 2016.

At the same time, the cost of accessing capable AI models has been falling rapidly.

This creates an unusual economic environment.

AI models may become cheaper for users while becoming more expensive for developers to build.

Revenue-sharing licenses could become one way for AI laboratories to recover development costs while still releasing downloadable model weights.

Alibaba already earns revenue when developers use Qwen through Alibaba Cloud.

A revenue-sharing model could allow the company to also earn money when large companies deploy Qwen independently, operate it through third-party cloud providers, or sell services built around the model.

Moonshot has already begun testing this approach with Kimi K3.

A New Business Model for Open AI?

The emerging trend could reshape what businesses mean when they describe an AI model as “open.”

Instead of choosing between completely free open-source models and fully closed commercial APIs, AI developers may increasingly adopt a hybrid model.

The weights remain downloadable.

Researchers and smaller developers can experiment freely.

But companies earning significant revenue from the technology may eventually be required to negotiate commercial agreements with the model developer.

This could create a new category of open-weight AI with enterprise monetization.

For developers, startups, cloud providers, and AI companies, the message is becoming clearer:

Open weights do not necessarily mean unlimited free commercial use.

As increasingly powerful AI models require billions of dollars in computing infrastructure, developers are experimenting with new ways to keep their models accessible while ensuring that companies making significant money from them also contribute financially.

Alibaba has not yet publicly announced the final license for its next Qwen model or confirmed the revenue-sharing percentage it may seek.

But if the reported plan moves forward, it could become another major experiment in how the global AI industry balances openness, accessibility, and commercial sustainability.