Ask a Western executive to name a Chinese robotics company and the answer comes back instantly: Unitree, the Hangzhou firm whose humanoid robots danced on the Spring Festival Gala and flip through every viral video feed. Ask where the supply chain that builds those robots lives, and the room goes quiet. The answer is not Hangzhou. It is Shenzhen, 1,000 kilometers to the south — a city that produces more robots than anywhere else in China and that almost nobody outside the industry is watching.
The Hangzhou illusion
Hangzhou's robotics fame is real but narrow. Unitree is a single-point star — brilliant, photogenic, easy to cover. When Hangzhou's 'six little dragons' went viral in early 2025, Chinese media looked for a counterweight and found one in Shenzhen's 'eight giants' (八大金刚). The framing stuck, but it missed the deeper point. Hangzhou has stars; Shenzhen has the system — the component suppliers, the contract manufacturers, the control-board engineers and the policy machinery that turns a sketch into a production line. The parts inside a headline-grabbing humanoid overwhelmingly come from the Pearl River Delta. The star performs on stage. The system lives in Shenzhen.
What Shenzhen actually built
The numbers are not small. In 2025 Shenzhen's robotics industry produced 242.6 billion yuan of output, up 20.6% year on year — the largest of any Chinese city, with the country's densest population of robotics firms. This did not happen by accident. Since 2022 the city has folded intelligent robotics into its '20+8' strategic industry-cluster program; in the July 2026 upgrade, robotics was named one of six 'strategic priority' clusters, a designation the government describes as marshaling the resources of the entire city to support it. In February 2025 Shenzhen announced a 10-billion-yuan AI and robotics industry fund, and a month later published a dedicated embodied-AI action plan targeting 100 billion yuan in related industry scale by 2027 and more than 1,200 cluster companies. The city subsidizes compute with 'training vouchers' worth up to 10 million yuan per firm and has committed to opening at least 50 real-world scenarios a year — factories, streets, hospitals, transit — so robots have somewhere to work.
But policy is not the product. The product is density. Inside the Robotics Valley in Nanshan district, a team can sketch a part in the morning and hold a prototype by the afternoon. The city's own industry bureau puts it plainly: components ordered in the morning are fabricated and delivered the same day. Within a 30-minute drive a humanoid builder can reach every supplier it needs — harmonic reducers, servo motors, tactile sensors, contract manufacturers willing to run small batches before any purchase order is raised. Engineers call it the 'half-hour supply circle.' It is the single most important fact about Chinese robotics, and it does not appear in any analyst report.
Eight companies, four ways to survive
The 'eight giants' are not one strategy repeated eight times. They are four distinct bets on how the humanoid race will be won.
The industrial incumbent — UBTECH (优必选). The first humanoid company to list on a major exchange, UBTECH bet on factories before homes. Its Walker S series already runs training shifts inside multiple Chinese car plants, and the Walker S2's ability to swap its own battery pack — a detail that sounds minor until you realize it removes the last reason to stop the line — is the kind of engineering call that only comes from watching a robot fail on a real factory floor. The bet: whoever wins automotive earns the right to everything else.
The volume player — Dobot (越疆). While the humanoid crowd chases backflips, Dobot has quietly shipped more than 100,000 collaborative robot arms, many to lines run by Fortune 500 manufacturers. The bet is less photogenic and arguably more durable: force-control and AI models fused into arms that already have paying customers across 3C electronics, automotive and semiconductor assembly. When humanoids finally reach those lines, they will be joining a workforce of Dobot arms that got there first.
The full-stack athletes — EngineAI (众擎) and LimX Dynamics (逐际动力). Both build full-size humanoids with in-house joint actuators and motion-control stacks, and both made their names on athletic demos: running, jumping and terrain traversal that would have been research-lab material three years ago. The athleticism is not the product. It is the proof of work for the actuator and control technology underneath, which is what these firms actually sell — to research labs, to integrators, and increasingly to factories that need a robot that can walk before it can work.
The picks-and-shovels layer — PaXini (帕西尼) and AI2Robotics (智平方). Every gold rush has its toolmakers. PaXini builds multi-dimensional tactile sensors, a 'skin' that lets a robot feel material, texture and grip force, closing the physical-interaction gap that vision-only systems cannot. AI2Robotics is building an end-to-end embodied foundation model, betting that the winning humanoid will carry the best general-purpose brain rather than the best body. Neither needs to win the humanoid race outright. They need the race to happen.
The institutions underneath — AIRS and the Robotics Valley itself. The Shenzhen Institute of Artificial Intelligence and Robotics for Society supplies the cluster's research substrate: algorithms, talent pipelines and the patient capital that venture funds cannot provide. The Valley in Nanshan is where the density becomes physical — upstream component makers, mid-stream integrators and downstream application teams stacked into the same few blocks, close enough that a design change propagates through the supply chain in days rather than quarters.
Why one afternoon beats a hundred reports
Here is what no report can give you. Standing in a Valley workshop, you can watch a procurement manager from a humanoid startup negotiate a custom reducer order with a supplier who shares the building. You can ask an engineer why they chose a harmonic drive over a cycloidal one and get an answer that names the supplier two floors down. You can hold a tactile sensor that survived a hundred thousand grip cycles next to one that did not. These are not details. They are the actual information content of the Shenzhen robotics cluster, and they are only available on site.
You can read every analyst report on Chinese robotics ever written. Or you can spend one afternoon in Nanshan and know more than all of them combined.
A two-day field itinerary inside the cluster
A typical program runs two days and five stops, all inside one compact cluster. The full eight-company lineup, including LimX Dynamics and AI2Robotics, is available on customized extensions of the schedule below.
Day 1 — tracing innovation to its source. The morning is at AIRS, one of Shenzhen's ten foundational research institutes: frontier robotics labs, applied-scenario research and an opening briefing on where the humanoid industry is heading. The afternoon moves to the Robotics Valley — the half-hour supply circle in person, a walk through the upstream-to-downstream showroom, and live demos from cluster companies such as EngineAI. The day closes at PaXini, China's tactile-sensing leader, for its multi-dimensional sensors, dexterous hands and humanoid platform.
Day 2 — from lab to production line. The morning is at UBTECH or Dobot: the listed leader's showroom, robots already deployed on real automotive and electronics lines, and a session on the trajectory of Chinese humanoids and their path into factory scenarios. The afternoon is at EngineAI for a close look at its full-size humanoid platform, followed by a closing roundtable where delegates and host companies explore concrete commercial cooperation.
China Innovation Visits arranges private, executive-level access to all eight companies and institutions named in this article — UBTECH, EngineAI, Dobot, PaXini, LimX Dynamics, AI2Robotics, AIRS and the Robotics Valley cluster — with technical briefings, engineering-team access and Mandarin-English interpretation included. If your next strategic decision depends on knowing where humanoid robotics actually stands, tell us your priorities and we will build the Shenzhen itinerary around them. Start at chinatechvisits.com/company-visits.
Questions delegations ask before they go
Do we need engineers in the delegation? No. The programme is built for executives and investors, with engineering teams available on site to answer at whatever depth you want.
Can we meet all eight companies? The two-day programme runs five stops. The full eight-company lineup, including LimX Dynamics and AI2Robotics, is available on customised extensions.
Is everything in one place? Yes. The cluster sits in Nanshan, and every supplier a humanoid builder needs — harmonic reducers, servo motors, tactile sensors, contract manufacturers — is within a 30-minute drive.
Is English interpretation provided? Yes. Mandarin-English interpretation is included across the technical briefings.
Arranging your Shenzhen humanoid-robotics programme
We arrange the visit for you end to end: the introduction to the host, the scheduling on the Chinese side and English interpretation on site. Tell us your dates, your delegation size and what you want to see, and we will confirm what can be arranged. Start at chinatechvisits.com/company-visits.