
San Francisco/BEIJING – Chinese technology giant Alibaba plans to ask major users of the next version of its Qwen open-source AI model for a share of revenue they make from the hotly anticipated offering, according to two people familiar with the company’s plans.
Like July’s blockbuster model release from Chinese AI start-up Moonshot, Alibaba’s Qwen3.8-Max model is an open-source, open-weight model, meaning that the underlying learned settings that allow developers to run or adapt the system are available for download. By contrast, US developers OpenAI, Anthropic and Alphabet’s Google have closed-source models.
Open source is often thought of as free, but tucked into the licensing terms of Moonshot’s Kimi K3 was a provision that requires anyone offering the model for sale as a service and generating more than US$20 million (S$25.7 million) in annual sales to work out a commercial agreement with Moonshot.
Alibaba plans to implement a similar measure for its open-source model next week, said the two people familiar with its strategy, which has not been reported previously. To date, Alibaba has charged developers for use of its models when hosted on its own cloud computing platform but allowed most of its open-source offerings to be used in customers’ own data centers without payment.
The plans show Chinese AI firms, which have shocked markets by releasing open-source models nearly as capable as those from OpenAI and Anthropic, are converging on a business model as they push to take market share from their US rivals.
Revenue-sharing deals
Deals under which US firms share revenue generated by Chinese models with the Chinese AI labs that created them are taking shape, even as the White House has accused Moonshot of stealing technology from Anthropic. Chinese officials have rejected that claim as unfounded.
The Kimi K3 license requires Moonshot’s partners to share revenue, according to the two people, with Moonshot requiring up to a 30 per cent revenue share, one of them said. Alibaba also plans to ask for a revenue share, both people said, but the rate remains unclear as the discussions are ongoing.
Moonshot did not respond to a request for comment. Chinasoft International, a Chinese IT services provider, last month disclosed a revenue-sharing agreement with Moonshot in a regulatory filing, without disclosing the percentage.
The Chinese AI firms are following a common playbook from Silicon Valley: Offer software at low or zero cost to start, charging for heavy commercial use and additional services.
“You pay for collaboration with these open-weight model labs to make sure that you’re optimizsing your deployment. You pay for getting early access for the next revision of the model,” said Paddy Srinivasan, chief executive officer of cloud computing firm DigitalOcean Holdings, one of several US firms offering Kimi K3 and other Chinese models.
He confirmed his company has a commercial agreement with Moonshot, but declined to discuss specifics.
“This is a tried and tested open-source ‘freemium’ model,” Srinivasan said.
Kimi K3 costs about a third of Anthropic’s Fable model, according to listed prices of input and output tokens for both models.
Dan Fu, vice president of kernels at Together AI, said companies like his that offer AI software make their money by optimizsing the service, such as better use of tokens, the building blocks of AI queries.
“At the application layer, there’s value out there for how you use it, how you actually get the models and the tokens to do something useful,” Fu said.
Open-source AI, meanwhile, is expanding, with US firms joining the fray.
Thinking Machines Lab, an AI start-up founded in 2025 by OpenAI’s former chief technology officer Mira Murati, in July released its first open-source model, and is widely expected to follow with more powerful ones.
“I don’t see a fundamental barrier” to powerful open-source US models, said Lin Qiao, CEO and co-founder of Silicon Valley-based Fireworks AI, who declined to discuss the company’s commercial arrangements with Moonshot. “We are really waiting for that to happen.” REUTERS



