World Robot Conference Opens: China Accounts for Over 90% of Global Humanoid Robot Shipments
Nashnova编辑部
The 2026 World Robot Conference opened in Beijing on August 19, with Chinese makers now accounting for over 90% of global humanoid-robot shipments and holding all top-five spots; yet a government-funded training-center loop is raising pointed questions about real demand and valuation logic.
Where does China's "90% share" actually come from?
Chinese manufacturers now account for over 90% of global humanoid-robot shipments and hold all five top spots.
Morgan Stanley raised its 2026 China shipment forecast from 28,000 to 50,000 units, citing stronger-than-expected purchases by local governments and commercial buyers.
This means → the shipment growth is backed by real orders, but *who* is placing them — and where the money originates — determines how much that "90%" is really worth.
Who is buying these robots? And where does the money end up?
According to the *Financial Times*, a significant share of revenue comes from government-backed robot training centers — the centers buy robots, collect training data, then sell the data back to the robot makers, forming a closed loop.
As of June, over 90 such centers were built or under construction nationwide. Shenzhen-based LeeJu Robotics disclosed that 45% of its flagship "Kuafu" humanoid's sales came from training centers. UBTech booked RMB 140 million in orders from government-backed centers; nearly three-quarters of Unitree's humanoid revenue in the first three quarters came from education and research users.
In plain terms = government funds the centers, centers buy the robots, data flows back to the makers — the money circles within the system, and genuine commercial customers (factories, logistics) remain a small slice.
What do investors think? Are some already heading for the exit?
A senior investor at a Beijing venture-capital firm told the *Financial Times*: "There is broad consensus among early investors that humanoid robots are approaching the peak of the hype cycle. We are looking for opportunities to sell down and exit."
Another investor was blunt: "If companies cannot prove robots can be deployed at scale in factories, investors will reassess valuations within the next year."
This means → the patience window in the investment community is roughly one year — not much time for makers to prove "the booth can become a factory floor."
What stands between the booth and the factory?
Nvidia's edge-AI marketing lead Chen Su noted that real-world demo data collection is too slow; the industry must rely on simulation-generated synthetic training data to build "world foundation models" — models that understand the laws of physics.
Intel's industrial-solutions director Ricky Watts cited a food-and-beverage client: an AI defect-detection system saved roughly $25,000 per year, but a single nine-month equipment outage cost about $100,000 — wiping out four years of savings and destroying the client's confidence in camera-based AI.
In plain terms = industrial robots must run in harsh conditions for fifteen to twenty years; safety-critical, latency-sensitive tasks cannot depend on the cloud. This real-world threshold is the key test of whether the sector's valuation logic holds up.
What is on display at the conference itself?
Over 300 companies are exhibiting more than 2,000 products, including over 150 first-time debuts.
Four themed halls cover the full supply chain: the Smart Innovation hall hosts Unitree, UBTech, Galbot, Fourier Intelligence, and LeeJu; the Smart Integration hall showcases dexterous hands, sensors, chips, joints, and other core components; the Smart Manufacturing hall spans industrial, service, and medical applications.
Unitree chairman Wang Xingxing will deliver a keynote on humanoid robotics over the next decade; Galbot founder Gao Jiyang will speak on how open-source ecosystems accelerate embodied AI — technology that gives AI a physical body to act in the real world.
Content is for reference only, not financial advice.