Funding heat in embodied intelligence shows no sign of cooling.
Gasgoo Embodied counts 56 disclosed financing events in China’s embodied-robotics and core-component space in August. The largest single round came from XPeng’s robot unit at more than US$900 million, a record for a single private equity round in China’s embodied-intelligence sector. Sharpa disclosed its first funding, above 4.5 billion yuan, at a value near 22 billion yuan.
PaXini, Zhongke Diwuji and Yuanli Wuxian also landed billion-yuan rounds in August. Just five firms took about 14 billion yuan between them, tightening the lead of the front runners.
Yet much of the capital is flowing early: nearly half of the 56 rounds are still seed, angel or pre-A, while growth rounds around series B are thinner. The pattern is hot at both ends and light in the middle.
From body to brain
The deeper signal is where the money goes. Almost half of August’s rounds went to ‘brain’ plays, both pure brain startups such as Moushen, Qianjie, Liangyuan and Yuanli, and full-stack players like XPeng, Sharpa, GAC-backed Huilun and Zhongke Diwuji that treat the brain as the moat.
Two years back the test was simple: could a robot walk, run and not fall. In 2026 the bar is whether a robot can perceive, react and finish a job in a messy real scene. The cap is brain generalisation, and that cannot be bought or rushed with parameters. It needs mountains of scene data fed over time.
Data infrastructure drew more than ten rounds in August, from Mifeng, Jianzhi, Current Robotics, Kaiwang, Lingyu, Aiai and Yuandian, each building the collection, labelling, training, testing and simulation loop as the ‘shovel seller’ of the sector.
China’s academy estimates that a ChatGPT moment for embodied base models needs on the order of ten million hours of real data, while the world now has only hundreds of thousands to a million hours, a gap of roughly tenfold. By end-June China had over 70 training grounds open and 46 more building.
The cool side showed in late August when Beijing’s first humanoid data centre stopped operating, less than a year and a half after opening. The shift was from lab tele-operation to real-scene remote operation, a signal that the centralised version 1.0 route is being retired, not the data thesis itself.
Dexterous operation is the other focus
In hardware, capital targeted the parts that decide dexterous work: joint modules, dexterous hands and force touch, the three gates of stable motion, a real grip and real feel.
Joint modules drew Yiuyou, Linghou, Nuoshi, Zhuoyu, Taoshi and Hengxuan. Joints are 35 to 60 per cent of a humanoid’s bill of materials, the thickest single module, with clear local-substitution room and sticky, co-defined supply relationships.
Dexterous hands took the lead from legs as mobility stopped being scarce. Sharpa, founded by the Hesai founding trio, drew above 4.5 billion yuan from Alibaba, Meituan, Tencent, JD and Sequoia China, and aims for the home from 2028 after a few real scenarios. Force sensing rose too, with LanDian Touch’s several-hundred-million-yuan series D led by GAC Capital and tactile bets in PaXini, Daimo, Moliang and Linggan.
The Unitree lesson
The private market’s stories still face the public market’s test, and August supplied a clear ruler. Unitree listed on the STAR Market on 19 August at 150.80 yuan, a 61 billion yuan issue value and 219 times earnings. It opened at 1,100 yuan, up about 629 per cent to a 444.9 billion yuan value, then fell to 523.68 yuan, a 211.8 billion yuan value, by 8 September, a drop of about 52 per cent from the peak.
The pullback reflects a high value and a tiny free float, a half-year result where recurrent profit slipped about 19 per cent as R&D rose about 152 per cent, and founder Wang Xingxing’s own warning that general robots need two to ten years for true generalization. Even the front-runner must be repriced, and the hundred-billion-yuan unicorns behind it will rethink their listing maths.
The 56 rounds look like a route vote: brains and data make robots clever, hands and joints make them useful. Unitree’s two-week slide stamped the word ‘delivery’ on the whole field.
Editor’s note: This is an adapted translation of the original Gasgoo Embodied report. It has been trimmed and restructured for readability for an international business audience.
Translated and adapted from Gasgoo Embodied (https://www.gasgoo.com/apps/50640d4b55d5cba175fb84f15d679f19/robot/news/70471481-8%E6%9C%88%E5%85%B7%E8%BA%AB%E6%9C%BA%E5%99%A8%E4%BA%BA%E8%9E%8D%E8%B5%84-%E5%85%B1%E8%AE%A156%E7%AC%94-5/).