By Sarah Xiao 
September was the month the mega-round music stopped in China’s embodied AI sector — and the month the money got more interesting.
According to tallies by Gaoshi Intelligence (盖世具身智能)， 49 robotics and embodied-AI companies disclosed funding in September, a count higher than most months this year. But only two rounds crossed the RMB 1 billion line: D-Robotics (地瓜机器人) with a US$400 million Series C, and Sudu Technology (苏度科技) with a new round above RMB 1 billion. Compare that with April and June, when ten RMB-100M-plus rounds closed in each month, and single rounds of US$450 million (Tastone Intelligent’s Pre-A, a sector record) and near-RMB 5 billion totals (Zhiqiping’s series) were making headlines.
Neither September mega-round went to a general-purpose humanoid maker. That is the real story.
Key takeaways
- Deal count stayed high; check sizes collapsed. September’s 49 deals vs. two above RMB 1B means capital is fragmenting, not leaving.
- Leaders have gone quiet for a reason: IPO windows. After Unitree’s August 19 STAR Market debut — China’s first listed humanoid maker — several头部 players paused large private rounds to protect valuation anchors.
- Money moved up the supply chain. Force- and tactile-sensing suppliers (Parseni, Xinjingcheng, Tujian, Qianjue, Juqiao, Boraizhi) all closed new rounds in September.
- Seed and angel rounds made up ~37% of September deals — the finest funding granularity of the year.
- Automotive-talent teams keep raising, because their track records are verifiable in a market that has become allergic to unverifiable claims.
Why the big rounds stopped
Two structural reasons, one on each side of the table.
On the company side: the head players are heading to IPO. Since Unitree’s listing, Zhiyuan Robotics, Leju Robotics, Galaxea (极佳视界) and Zhiqiping (智平方) are all reported to be in or approaching listing processes. Companies in an IPO window routinely freeze large equity rounds — new money at new prices destabilizes the valuation anchor and complicates the issuance review. Listing itself is also simply a bigger, cheaper financing channel than another private round.
On the capital side: valuations outran exit paths. As of mid-October, at least ten private or pre-IPO companies are reported at or above RMB 20 billion valuations — Zhiyuan, Galaxea, Zivariant (自变量)， Zhiqiping, Galbot (银河通用)， Xinghaitu, Qianxun Intelligent, Lingxin Whisker, Sharpa and Sudu among them. At those prices the question is no longer whether a fund believes the thesis; it is whether the fund can write the check. Meanwhile, September brought repeated reports of tighter IPO review for hard-tech issuers — sharper scrutiny of valuation reasonableness, revenue authenticity, and whether government-subsidized projects represent durable demand. Betting another nine figures at the top of that curve stopped being a value play and became a pure risk trade.
Where the money went instead
Fragmented, earlier, and further upstream.
Earlier: with seed and angel rounds at roughly 37% of September’s disclosed deals, funds are buying tickets across more, smaller positions. Sidu Technology raised three rounds in a month; Sainade and Westlake Robotics each closed four in six months.
Upstream: the clear cluster is force and touch. Six tactile/force-sensing suppliers — Parseni (帕西尼)， Xinjingcheng (鑫精诚)， Tujian (途见)， Qianjue Robotics (千觉)， Juqiao Industrial (矩侨工业) and Boraizhi (柏锐智芯) — raised new money in a single month. Capital that spent 2025 buying “brains” is now buying the fingers.
Toward automotive DNA: at least 7 of September’s 49 funded companies have explicit automotive or autonomous-driving roots. D-Robotics’ leadership comes from Horizon Robotics; Xieyue Intelligence was founded by Li Auto’s former AI chief scientist Chen Wei; Chaowei Powerll’s founder Yu Jie ran Huawei’s commercial-vehicle autonomous driving unit; Xiyuan Robotics is controlled by RoboSense’s founder. Per Gaoshi’s running count, roughly 40 senior AV executives and engineers have crossed into embodied AI since 2023, across 20+ startups — with Horizon, Li Auto and Huawei’s car BU the three densest talent pools.
The logic is risk-adjusted, not fashionable. In a market burned by demo-driven claims, a team whose end-to-end stack actually shipped on vehicles can be diligenced: did it deploy, at what volume, is the data loop closed? Chen Tao Capital’s estimate that only about 20 people in China combine mass-production delivery, end-to-end model training, massive data handling and cluster-scale operations explains both the scarcity premium and the oversubscription. And autonomous driving remains the only domain that has walked the full loop — algorithm, training, mass delivery, data feedback — that embodied AI now needs.
The demand side refuses to keep pace
IDC data puts first-half 2026 global humanoid shipments near 25,000 units (+432% YoY) and market size above US$740 million (+322.7%), with China at roughly 78% of units. But the mix undercuts the headline: per Gaoshi partner Wang Xianbin, entertainment, data collection and research uses still account for 35%, 20% and 10% of shipments respectively, against 15–20% combined for 3C, automotive and logistics. Even an optimistic full-year forecast of 60,000 global units (some see 100,000) cannot support the implied growth in a RMB-20-billion valuation.
Hence the bubble debate. Lenovo Capital’s Wang Guangxi: “There is definitely a bubble — these valuation moves don’t follow tech or business logic.” Zhu Xiaohu of GSR Ventures went further: “China has 200-plus robot companies. Which one can’t dance? Which can’t box? This market is too small; most won’t survive ten years.” Ecovacs founder Qian Dongqi expects a visible squeeze between listed valuations and actual business closures.
The bull rebuttal is the infrastructure argument — the dot-com fiber analogy: someone has to overbuild the rails before the real companies arrive. Both can be true, which is why “structural bubble” is the most defensible read. The froth concentrates in general humanoids and “brains” with similar roadmaps and thin field validation; key components, sensors and companies with real industrial demand remain early-stage and comparatively underpriced. New Dimension Capital’s Zhang Chi expects a 2026 shakeout that leaves 10–20 survivors among 100-plus funded humanoid makers; Matrix Superintelligence’s Zhang Haixing puts the window at 18 months.
What component buyers should take from this
- Supplier risk is now survivability risk. Early-stage sensor money means more choice and more mortality. Qualify tactile and force-sensing suppliers on cash runway, reference installs and second-source plans — not just spec sheets.
- IPO-bound OEMs will prioritize industrial demand. Companies heading to listing need revenue quality; your repeat orders matter more to them than a demo contract. It is a good window to negotiate support terms.
- Watch the sensing cluster. Six funded force/tactile suppliers in one month means capacity and price competition ahead in exactly the components dexterous hands need — a rare buyer’s market forming before the volume arrives.
- Discount announcement volumes. Strategic cooperation counts and cumulative round figures are not shipments or cash. The two September mega-rounds (compute platforms, reusable embodied platforms, not humanoids) tell you where sophisticated money thinks the next durable margin sits.
September didn’t cool the sector. It repriced it. The winners of the next eighteen months will be the companies converting orders — not rounds — into cash flow, and the suppliers positioned underneath them.