Alibaba shifts Qwen strategy: focus moves from open source to paid AI services

Alibaba is shifting its approach to Qwen development: the company no longer wants to compete solely on the quality of open-source models and is accelerating their transition to paid cloud services. Formally, the Qwen family is not disappearing, but inside Alibaba, priority is shifting from the research race and benchmarks to products that can generate revenue right now.

Shift to Monetization

In the first years of the generative AI boom, major companies eagerly invested in scaling models, even though direct economics remained unclear. Now Alibaba has a different focus: management wants the AI direction to work more strongly for the cloud business and stop being a showcase of technological achievements. Company head Eddie Wu publicly promotes the MaaS model — where customers pay for model usage based on actual load. The problem is that despite Qwen's notoriety, this format has not yet generated noticeable revenue for Alibaba, so the company needs a more rigorous commercial framework.

Team Reshuffles

Two notable figures have already left the Qwen team, people who largely defined the previous direction. This concerns former Qwen technical director Lin Junyan and researcher Hui Binyuan, who specialized in code generation and writing tasks. These people were associated with Qwen's engineering vector and its reputation among developers.

These changes look not like routine personnel rotation, but as a signal that Alibaba is reorganizing priorities around a more applied and profitable strategy. The new management center is also structured for this task. Eddie Wu personally heads the AI strategy committee, and operational management of the direction is handed to Zhou Jingzhen, who was previously Alibaba Cloud's technical director.

His main task is to link model development with cloud business metrics. This is a notable shift: previously the team paid much attention to model speed and results in industry benchmarks, but now they expect not only strong science, but clear commercial results.

What's Changing in Qwen

Alibaba is not abandoning AI altogether and is not burying Qwen as a brand, but is changing the packaging and business model. The company has already begun releasing closed-source models, access to which customers receive through Alibaba Cloud infrastructure. At the same time, it is not planning to completely move away from open source: some models will remain open for certain scenarios and ecosystem tasks. That is, instead of the previous bet almost exclusively on popularity among developers, Alibaba is building a mixed scheme: open models for reach and closed ones for direct monetization.

For Alibaba, this is a difficult turnaround precisely because Qwen gained weight thanks to open distribution. Open models helped the company gather a wide community of developers around itself and strengthen its reputation in the global AI environment, but popularity alone does not guarantee cash flow. The more the ecosystem became accustomed to perceiving Qwen as an accessible tool, the harder it becomes to explain what exactly the customer should pay for in the cloud. Hence the bet on closed versions, service scenarios, and tighter integration with Alibaba Cloud infrastructure.

• Token Hub division launched to accelerate AI commercialization
• Eddie Wu personally oversees the new AI strategy committee
• Zhou Jingzhen is responsible for linking models to cloud revenue
• Alibaba has already presented several closed-source models
• Open-source line will be preserved for some scenarios

What This Means

For the market, this is yet another sign that the era of "first growth, then money" in generative AI is ending. Even companies with strong open-source models are now forced to prove that their developments can generate stable revenue through cloud, APIs, and agent services. For Alibaba, such a turnaround is especially important: it already has a recognizable AI brand and a large developer base, but the next stage will depend not on the number of Qwen downloads, but on how successfully the company converts interest in models into a paying business.