Alibaba is considering a new commercial strategy for its Qwen AI models that could change how large companies use its open-weight technology.
The company is reportedly planning to introduce revenue-sharing requirements for certain commercial users of its next Qwen model. Under the proposed approach, larger businesses that generate revenue by providing the model as a service could be required to negotiate a commercial agreement with Alibaba.
The specific percentage Alibaba intends to seek has not yet been finalised.
A different approach to open AI models
Alibaba has historically allowed users to download and deploy its open-weight models on their own infrastructure without licensing fees, while charging developers who access the models through its cloud platform.
The proposed terms would represent a shift from that approach.
The company’s current Qwen3 open-weight models are released under the Apache 2.0 licence, which permits commercial use, modification, and redistribution as long as the licence conditions are followed.
The distinction between open-source and open-weight AI is important here. Open-weight models make their trained parameters available for download, but that does not necessarily mean every component of the system is openly accessible or that commercial use comes without restrictions.
The Open Source Initiative’s definition of open-source AI goes further, requiring users to have the ability to use, study, modify, and share the system while also providing access to information such as training data, relevant code, and model parameters.
Alibaba is not alone
Alibaba’s potential strategy follows a broader trend among Chinese AI companies experimenting with ways to monetize open-weight models.
Moonshot, the developer behind Kimi K3, has introduced commercial conditions for companies operating Model-as-a-Service businesses above certain revenue thresholds.
Under Kimi K3’s licence, companies must enter into a separate agreement with Moonshot when the combined revenue of the company and its affiliates exceeds $20 million over any consecutive 12-month period. The requirement also applies to commercial use of derivative models.
Large consumer-facing deployments can face additional requirements. Commercial products exceeding 100 million monthly active users or $20 million in monthly revenue must prominently display the Kimi K3 name, subject to certain exemptions.
Moonshot’s commercial agreements can also include revenue sharing. One source cited in the report said the company can require partners to share as much as 30% of relevant revenue.
Running open models still costs money
Making an AI model available for download does not eliminate the infrastructure costs associated with operating it.
Large models require significant computing resources, particularly when businesses deploy them at scale.
Kimi K3, for example, contains 2.8 trillion total parameters and 104 billion activated parameters. Its mixture-of-experts architecture contains 896 experts, with 16 selected for each token.
Alibaba is pursuing a similar architecture with Qwen3.8-Max. According to the report, the model contains roughly 2.4 trillion parameters while activating approximately 95 billion parameters for each request.
Mixture-of-experts systems can improve efficiency by activating only a portion of a model’s available experts for each token rather than running the entire model every time.
That creates opportunities for companies beyond the model developer itself. Cloud providers can earn money from hosting and inference, while AI infrastructure companies can generate revenue through deployment, optimisation, and performance improvements.
Open models can become a freemium business
The emerging model could resemble a freemium strategy.
Businesses may receive access to AI model weights at little or no initial cost, while companies operating the models commercially at significant scale could eventually pay for licensing, infrastructure support, technical services, or access to future releases.
DigitalOcean CEO Paddy Srinivasan described the approach in those terms, noting that companies can start with relatively low-cost access before paying for additional commercial services or capabilities.
This could give AI developers a way to preserve the advantages of open-weight distribution while still capturing some of the economic value created when their models become part of large commercial products.
Training costs are rising
The economics become more complicated when the cost of developing increasingly capable models is considered.
Research involving Epoch AI and Stanford researchers has estimated that the cost of the most compute-intensive AI training runs has increased by approximately 2.4 times per year since 2016.
At the same time, the cost of accessing models at a particular level of capability has fallen significantly, creating pressure for model developers to find sustainable ways to monetize increasingly expensive systems.
Alibaba already generates revenue from developers that access Qwen models through Alibaba Cloud. Revenue-sharing agreements could provide another source of income from businesses that independently deploy Qwen on their own infrastructure or through third-party platforms.
US and Chinese AI strategies are diverging
The development comes amid broader competition between Chinese and US AI companies.
Chinese developers have increasingly embraced models with downloadable weights, while major US companies such as OpenAI, Anthropic, and Google primarily distribute their leading commercial models through hosted services and more closed systems.
That does not mean US companies are abandoning open models. Thinking Machines Lab, founded by former OpenAI CTO Mira Murati, has also released an open-source model.
Fireworks AI CEO Lin Qiao has argued that there is no fundamental technical obstacle preventing US developers from releasing more capable open-source models.
The difference increasingly appears to be about business strategy rather than simply technical capability.
Alibaba has yet to finalise the terms
Alibaba has not publicly announced the final licence for its next Qwen model or confirmed the percentage of revenue it would seek from large commercial users.
If the company proceeds with the plan, however, it could signal an important evolution in the economics of open AI.
Rather than treating downloadable model weights as the end of the commercial relationship, AI developers could increasingly use open distribution as the starting point for monetizing deployment, infrastructure, technical support, and large-scale commercial applications.
For companies building products on top of open-weight models, that could make licensing terms an increasingly important part of the AI infrastructure equation.
Source: https://www.artificialintelligence-news.com/news/alibaba-qwen-open-source-ai-revenue-sharing/
