Ant Group’s embodied-AI subsidiary, Shanghai Ant Lingbo Technology, has opened its first financing round with a target of 1.5 billion yuan and plans a second close by the end of the year, in what would be one of the fastest fundraising processes in China’s embodied-intelligence sector.
The brain, not the body
Lingbo was founded in December 2024 as a wholly owned Ant Group unit with 100 million yuan of registered capital. Its thesis mirrors that of US firm Physical Intelligence: do not build the hardware, build the general intelligence that any robot can run. In July it released the LingBot 2.0 model family, a full stack covering perception, understanding, decision, action generation and world prediction, including an embodied base model that adapts to more than 20 robot configurations and a spatial-native vision base model.
Capital is already pricing the brain
The market has rewarded this route. Physical Intelligence, which ships almost no physical robots, is in talks to raise roughly 1 billion dollars at a valuation above 11 billion. Skild AI closed about 1.4 billion dollars in January at a value above 14 billion, tripling in seven months. Even hardware-led Unitree is steering 2.02 billion of its 4.02 billion yuan IPO raise, about 48 per cent, into robot-model research. The consensus is that hardware standardises and can be procured, but world models, action models and interaction data are the core assets that cannot be bought.
Three hurdles
Analysts caution that Lingbo’s long-term value rests on more than model benchmarks. Its models must keep performing in messy real-world conditions, not just polished demos. It needs a scaled deployment loop where robot-environment interaction data continuously feeds model improvement. And it must graduate from an internal Ant team into a platform serving external hardware makers, building an independent commercial path rather than leaning on group resources. With global high-quality robot-interaction data still scarce and commercial cases rare, the race is far from settled.
Read the original report (OFweek Robotics)
Translated and adapted from OFweek Robotics (robot.ofweek.com).