Why a Former Alibaba Star Is Launching a $2B AI Lab Focused on World Models and Embodied Intelligence

From Alibaba to a New AI Lab

According to The Information, former Alibaba Qwen chief Lin Junyang, a 2019 Peking University linguistics master who rose from P6 to P10 at Alibaba’s DAMO Academy, is seeking a $2 billion valuation for a new AI laboratory. Investors such as Gaorong Capital and Sequoia China are in talks to fund the venture, which will pivot toward world models and embodied brains.

Open‑source Success of Qwen

Lin oversaw the full‑size open‑source release of Qwen at the end of 2022, giving Alibaba a genuine technical showcase in the global developer community. By the end of 2025, Qwen downloads on Hugging Face exceeded 600 million, spawning over 170 000 derivative models; early 2026 saw downloads surpass 1 billion with more than 200 000 derivatives, cementing its position as the leading open‑source large model worldwide.

Organizational Split and Departure

On 3 March 2026, Alibaba Cloud CTO Zhou Jingren reorganized the Qwen team, breaking the vertically integrated structure into horizontal groups for pre‑training, fine‑tuning, text, vision, and speech. The vertical model favored rapid iteration, while the horizontal split emphasized resource reuse and large‑scale collaboration. One day later, Lin announced his resignation; Alibaba held an emergency all‑hands meeting where executives Wu Yongming, Jiang Fang and Zhou Jingren attempted to calm the team.

Post‑departure Commercial Push

After Lin’s exit, Alibaba accelerated Qwen’s commercial integration, embedding it deeply into Taobao, Tmall, and DingTalk. During the Spring Festival, the Qwen app peaked at 73.52 million daily calls but quickly fell back to around 40 million, indicating modest retention.

Market Landscape and Funding Race

The large‑model arena now features giants ByteDance, Alibaba, and DeepSeek. World models and embodied intelligence are emerging as non‑consensus, high‑risk tracks attractive to entrepreneurs. Capital is pouring in: Fei‑Fei Li’s World Labs reached a $5 billion valuation with a $1 billion round in February 2026; AMI Labs secured a $1.03 billion seed round without a product; Genesis AI raised $105 million for a general‑purpose robot model; Chinese firms such as Jiji Vision and Qianxun AI raised tens of billions in recent months. Despite the funding frenzy, the Ministry of Industry and Information Technology reports over 140 domestic humanoid robot manufacturers, only a handful achieving thousand‑unit production and stable 24/7 operation.

Technical Hurdles of Embodied AI

Most players simply attach a vision module and an action library to an open‑source language model, claiming end‑to‑end embodied capability. However, language models generate text probabilistically, while the physical world obeys deterministic physics and real‑time feedback. Understanding a robot’s dynamics and executing long‑horizon tasks therefore requires more than symbolic reasoning.

Lin’s New Venture Outlook

Lin’s new team already includes core members from ByteDance, Tencent, and overseas backgrounds. The $2 billion valuation bets not on immediate technical maturity but on the market’s hunger for a fresh narrative after large‑model saturation. Lin’s proven ability to engineer and open‑source successful models adds a premium to the valuation.

Reference: https://www.theinformation.com/articles/former-alibaba-star-researcher-starts-new-ai-lab-seeks-2-billion-valuation