KunlunXing昆仑行
Latest3 rounds of financing (announced Jun 2026)
Over $1B (unicorn; exact figure undisclosed)
Beijing KunlunXing Robotics Technology Co., Ltd.
Completed 3 funding rounds totaling billions of RMB within 90 days of registration, a 2026 record for 0-to-1 embodied AI startups in China; a dual-system robot intelligence architecture centered on the Kunlun World Model (KWM), taking a physical-causality route
数据截至2026-08-19
About
Beijing KunlunXing Robotics Technology Co., Ltd. was registered on March 16, 2026 and settled in the Beijing Economic-Technological Development Area (E-Town). It was co-founded by Ren Geng (founder and CEO) - a former Huawei country CEO and president of Alibaba Cloud China - and Lang Xianpeng (co-founder), former senior vice president of Li Auto and head of its autonomous driving. The company builds the Kunlun World Model (KWM) with physical causality as its first principle, constructing a dual-system intelligence architecture on top of mainstream VLA; it originated the 'data compilation' paradigm and the industry's first counterfactual evaluation system. It self-develops high power-density drive motors, precision planetary gearboxes, and integrated joint modules, pursuing a trinity of hardcore robot bodies, embodied AI brains, and commercialized volume delivery, with its general-purpose humanoid robot benchmarked against Tesla. Within 90 days of registration it completed 3 funding rounds totaling billions of RMB, reaching a valuation above $1B and unicorn status - a 2026 record for 0-to-1 embodied AI startups in China. Investors include Hillhouse, Gaorong, CAS Star, Sinovation Ventures, and C&D Capital, with first-round investors following on in all three rounds. It explicitly forgoes prototype-only R&D and expects to land first in industrial manufacturing scenarios.
Technical Approach
The KWM world model takes physical causality as its first principle, introducing physical-causal modeling into a dual-system architecture (VLA + world model fusion) on top of mainstream VLA; the model uses an MoT multi-expert network architecture - one model, many functions; one brain, many forms - supporting cross-embodiment transfer and software-defined hardware. Its data system upgrades from the 'data loop' to 'data compilation', with the industry's first counterfactual evaluation system; it self-develops motors, reducers, joint modules, and low-level motion control algorithms, and explicitly forgoes prototype-only R&D.
Commercialization(5)
- From day one the company has anchored on engineering mass production and scenario-based commercialization, explicitly forgoing prototype-only R&D (QbitAI interview, 2026-08-01)
- Landed in Beijing E-Town on Jun 9, 2026, with the core team fully in place within two weeks, and completed corporate structure, large-scale talent build-up, and R&D system setup within 2 months
- Represents Beijing's new dual-drive paradigm of Haidian R&D plus E-Town industrialization for embodied AI
- Lang Xianpeng: current funding is already enough to bring the first product to mass production and market, after which the company will enter a healthy state of positive cash generation
- Continuously accumulates interaction data through real ToB device delivery, building a data flywheel and commercial loop; expected to land first in industrial manufacturing scenarios
Position & Strengths(5)
- A top founding team: Li Auto smart-driving chief Lang Xianpeng plus former Alibaba Cloud China president Ren Geng, a rare technology-plus-business combination praised by Hillhouse Ventures as scarce
- A differentiated physical-causality route: not chasing the mainstream VLA imitation route; centers on a physical-causal world model with the original 'data compilation' paradigm and counterfactual evaluation system
- Strong capital endorsement: 3 rounds totaling billions of RMB within 90 days at a valuation over $1B, setting the industry's fastest unicorn record
- Mass-production DNA: Lang Xianpeng delivered advanced smart driving solutions for 1.5 million vehicles; the team carries million-unit production experience
- 'Body + brain' dual drive: benchmarked against Tesla's humanoid robot, fully in-house, covering both body and brain