Alibaba’s Qwen AI Division Hemorrhages Talent as Three Senior Leaders Exit in 90 Days
Just 24 hours after shipping its flagship Qwen 3.5 models to global acclaim, Alibaba's AI unit lost its tech lead and two researchers—raising existential questions about open-source strategy versus commercial pressure.
Alibaba’s Qwen AI division has lost three senior executives in the first quarter of 2026, including tech lead Junyang Lin, who stepped down March 3 just one day after the team released its Qwen 3.5 models to praise from Elon Musk. The departures, which colleague testimony suggests were not voluntary, come as Qwen’s mobile app reached 203 million monthly active users in February—a sevenfold increase from January—and as Alibaba’s models surpassed 1 billion downloads.
Lin announced on X that he was stepping down as tech lead for Qwen, writing only “me stepping down. bye my beloved qwen.” A Qwen contributor hinted Lin’s departure wasn’t his choice, writing “I know leaving wasn’t your choice” in a post that characterized the team as “heartbroken” hours after celebrating the Qwen 3.5 launch together.
The exodus extends beyond Lin. Yu Bowen, who headed post-training for Qwen, also resigned on March 4, according to Chinese outlet LatePost. Staff research scientist Hui Binyuan, focused on coding, departed in January. Researcher Kaixin Li announced departure around the same time, while Binyuan Hui’s X profile now says “former MTS at Qwen”.
## The Strategic Fault Line
The timing crystallizes a conflict between Qwen’s open-source ethos and Alibaba’s commercial imperatives. Lin steered Qwen from a nascent lab project to a global powerhouse with over 600 million downloads, building influence through radical transparency: the project recorded over 600 million downloads and inspired more than 170,000 derivative models on Hugging Face, exceeding Meta’s Llama.
But analysts note Qwen has contradictions between its commercialization and technical approach—it has many followers in the global open-source community, but this approach has restricted Alibaba’s ability to directly generate revenue through APIs, according to Yicai Global.
“Replace the excellent leader with a non-core people from Google Gemini, driven by DAU metrics. If you judge foundation model teams like consumer apps, don’t be surprised when the innovation curve flattens.”
— Xinyu Yang, DeepSeek Researcher
Lin’s departure is closely related to organizational adjustments at the Qwen team—Alibaba’s Tongyi Laboratory plans to split Qwen from the “vertical integration” system, according to Yicai Global. Alibaba has recently consolidated its AI efforts into the “Qwen C-end Business Group,” merging its model labs with consumer hardware teams, per VentureBeat.
## China’s AI Talent War Intensifies
The departures unfold against ferocious competition for AI talent across China. ByteDance increased its bonus pool by 35% and allocated 150% more for future salary increases, while Tencent has lured several researchers from rivals by offering to double their existing salaries, according to South China Morning Post. The supply-to-demand ratio for AI talent in China stands at a mere 0.5, meaning there is just one suitable candidate for every two AI posts, per talent firm Hudson RPO.
Wedbush analyst Dan Ives said the exodus indicates underlying structural issues, and AI talent in China might be looking to be more free or better compensated amid export controls and increasing regulatory pressure, reported Techi.
Lin became the youngest P10-level tech lead at Alibaba last year. Before Qwen, he contributed to Alibaba projects including the mixture-of-experts model M6, the ICML 2022 paper OFA on multimodal pretraining, and Chinese-CLIP, a widely used Chinese-language vision-language model. At a January 2026 Tsinghua University AI summit, he spoke candidly about China’s compute disadvantages compared to the US, by one to two orders of magnitude.
The market reacted immediately. Alibaba shares slid 5.1% to HK$127.90 on March 4, among the biggest weights on the Hang Seng, according to Investing.com. The decline came despite Qwen’s mobile app surging to 203 million monthly active users in February from 31.05 million in January, now ranking third globally behind OpenAI’s ChatGPT and ByteDance’s Doubao.
## The Open Source Dilemma
Lin was the primary bridge between China’s deep engineering talent and the Western open-source ecosystem—without his advocacy, there are fears the project will retreat into a “walled garden” strategy, reports VentureBeat. Industry analysts warn that as Alibaba pushes to meet investor demands for revenue growth, the “open” in Qwen’s open-weight models may become a secondary priority—enterprises relying on Apache 2.0-licensed Qwen models now face the possibility that future flagships will be locked behind paid, proprietary APIs.
- Three senior Qwen leaders departed within 90 days, including tech lead Junyang Lin hours after the Qwen 3.5 launch
- Colleague statements suggest Lin’s exit was involuntary, tied to organizational restructuring prioritizing commercialization
- Qwen reached 203M monthly active users and 1B+ downloads, but open-source strategy conflicts with Alibaba’s monetization goals
- China’s AI talent war sees ByteDance raising bonuses 35% and Tencent offering double salaries to poach researchers
- Alibaba shares fell 4% on the news, with analysts citing structural tensions between research culture and business metrics
For the 90,000+ enterprises currently deploying Qwen via DingTalk or Alibaba Cloud, the leadership vacuum creates a crisis of confidence. Research scientist Wenting Zhao described Lin’s departure as “the end of an era,” thanking him for helping drive the project’s advances in open source AI, according to TechCrunch.
## What to Watch
Alibaba reports fiscal Q3 earnings March 5—executives face pointed questions about AI leadership continuity and retention strategy. The company has not announced Lin’s replacement or addressed the structural tensions driving the exits. Watch whether Qwen maintains its aggressive release cadence: Alibaba has released more than 400 open-source Qwen models since 2023, a tempo that demands stable technical leadership.
Monitor derivative model creation on Hugging Face and ModelScope—velocity declines would signal developer confidence erosion. ByteDance’s Doubao holds 250 million monthly actives versus Qwen’s 203 million; talent stability will determine whether Alibaba can close that gap or watch its hard-won market position erode. The broader question: can any AI lab sustain open-source leadership under quarterly earnings pressure, or does commercialization inevitably devour the culture that built breakthrough technology?