AI & TechArtificial IntelligenceNewswireStartupsTechnology

Qwen’s creator launches new lab backed by Shanghai government

▼ Summary

– Junyang Lin, former lead of Alibaba’s Qwen models, launched Pragmatik Labs in Shanghai, funded by Gaorong Ventures, HSG, Tencent, and the Shanghai Future Industries Fund, a $1.4bn government-backed entity that was not reported by major media.
– Lin’s background is in English literature and linguistics, not computer science; the lab’s name references pragmatics, the study of meaning in context, aligning with his focus on real-world functionality over benchmarks.
– Pragmatik works on digital agents for knowledge work, physical agents for real environments, frontier research, and feedback-driven products, with physical agents requiring costly perception and spatial reasoning capabilities.
– Lin left Alibaba in March 2025 after contributing to Qwen’s early models and warning about China-US AI gaps; Alibaba has since released Qwen 3.8-Max and shifted back toward open weights.
– The startup has no model, product, or demo yet; funding prices Lin’s reputation and research agenda, with the first technical output determining if the reported $2bn valuation was justified.

The researcher who built Alibaba’s flagship Qwen AI models has struck out on his own. Junyang Lin unveiled Pragmatik Labs on X Tuesday evening, exactly five months after resigning from the e-commerce giant. The new venture, headquartered in Shanghai, is focused on what Lin describes as next-generation agents designed to operate across both digital and physical environments. Gaorong Ventures and HSG, formerly known as Sequoia China, co-led the funding round, with Tencent also participating. Yet the most significant backer went largely unnoticed by major outlets.

The investor everyone missed

In his announcement, Lin thanked the Shanghai Engine Fund, using its Chinese name, which translates to the Shanghai Future Industries Fund. This is not a typical private investor. The fund commands a massive 10 billion yuan war chest, roughly $1.4 billion, and is financed entirely by the Shanghai municipal government. Launched in September 2024 with a 15-year mandate that can be extended by three years, the fund targets six designated future industries and deploys capital both directly and via sub-funds.

Mainstream coverage overlooked this detail. The Information cited Gaorong and HongShan, while Bloomberg listed Gaorong, HSG, and Tencent. The state’s involvement only surfaced in Lin’s own social media post.

Why the name is the thesis

Lin’s path to AI was unconventional. He earned a bachelor’s degree in English literature and a master’s in linguistics from Peking University, later leading a research group at the university’s Institute of Computational Linguistics. He joined Alibaba’s DAMO Academy in 2019, initially working on language generation for Taobao search before Qwen emerged. The lab’s name traces back to his academic roots.

“I studied linguistics because my friend recommended pragmatics to me,” he wrote. “The name just means being back to where it happens.”

Pragmatics, the study of meaning in context, neatly describes what an intelligent agent must master. Lin also invoked pragmaticism, the philosophy of evaluating systems by real-world effectiveness rather than benchmark scores.

What he is actually building

Pragmatik’s website outlines four workstreams: digital agents for knowledge work, physical agents for real environments, frontier research, and products refined through user feedback. The physical component carries the heavy costs. Digital agents require planning, tool use, and memory; physical agents demand perception, spatial reasoning, and actuated machinery. Pursuing both simultaneously represents a far larger financial commitment than building an app atop someone else’s model. Lin has not specified what the physical agents entail, though the term typically points toward robotics. China already dominates this space, shipping nearly every humanoid robot sold globally, placing the entire supply chain on Lin’s doorstep.

The exit that started it

Lin departed Alibaba on 4 March with a terse post: “Bye my beloved Qwen.” Two other senior researchers had left around the same period, and Lin had publicly voiced concerns about the widening gap between Chinese and American AI capabilities. His Qwen pedigree is substantial. He co-authored the original 2023 technical report and served as a core contributor to Qwen2.5, which became the foundation of Alibaba’s AI product line. Alibaba has continued advancing, releasing Qwen 3.8-Max this month, its largest model yet, with developer customization options. Lin also championed open weights, a stance that appeared precarious at Alibaba during his departure. The company has since pivoted back toward openness while separately exploring monetization for its heaviest open-model users.

A wave, not a one-off

Lin represents a broader trend. Elite researchers are leaving major labs worldwide, and China has developed its own cohort of what investors call neolabs. Capital has followed swiftly. Chinese AI startups have experienced a funding surge producing unicorns at remarkable speed. The strategic backdrop involves open weights, with cheap Chinese models creating what one analysis termed a death zone for American competitors attempting to sell access. Agents represent the next battleground. Alibaba itself has been steering models toward agents and robots, the same bet Lin is now making independently.

What is missing

Pragmatik currently has no model, product, or demo. The company offers a website, a mission statement, and an email address soliciting serious inquiries. Lin has not disclosed a valuation. The Information reported in May that he sought several hundred million dollars at a $2 billion valuation, though neither Lin nor his backers have confirmed figures since. Gaorong confirmed the funding without elaboration. HSG and Tencent did not respond to Bloomberg inquiries, and Lin declined comment.

This round effectively prices a reputation. It purchases the Qwen name, the talent Lin can recruit, and a research agenda spanning software and machinery, all before a single model has been released. Whether that price was reasonable or excessive will only be determined by the first technical output.

(Source: The Next Web)

Topics

ai startup funding 95% ai talent exodus 93% Agentic AI 92% open weights models 90% china ai ecosystem 89% ai research leadership 86% humanoid robotics 85% ai model development 84% startup valuation 82% government investment funds 80%