WRC 2026: Chinese Humanoid Robots Hit 97% of Global Shipments - From Dancing Demos to Real Work

Forum topic · QianXun · 2026-08-23

Summary

At WRC 2026, new data showed China's humanoid robot industry reached an inflection point. In the first half of 2026, China's robot industry scale hit 165.5 billion yuan, up 24.5% year-over-year, and Chinese humanoid robot shipments accounted for roughly 97% of global volume, according to a CAICT and China Electronics Society report corroborated by Counterpoint Research (~22,000 units globally, up nearly 300%). Unitree shipped over 7,000 units, AgiBot about 9,700, and Galbot over 1,100, with the top five vendors holding ~86% of the market. However, application breakdown reveals only a small fraction work in real manufacturing: entertainment performances account for 33.6% and research/data production 27%. Key hidden shifts include a 10-billion-yuan robotics VC fund by Dreame, a MIIT-backed real-world training program targeting 100+ high-value scenarios, and a split between real-data-driven and world-model/simulation-driven training approaches. Industry leaders caution true industrialization requires robots handling ~80% of tasks in unfamiliar environments, expected between 2026 and 2028.

WRC 2026 + Chinese Humanoid Robots at 97% of Global Shipments: The Inflection Point from "Dancing" to "Working"

On August 22, a Xinhua report put a striking number before the public: in the first half of 2026, China's robotics industry reached 165.5 billion yuan in scale, up 24.5% year-over-year. Under the same scope, Chinese humanoid robot shipments accounted for 97% of the global total, and the Ministry of Industry and Information Technology (MIIT) launched a "Humanoid Robot Real-World Training Special Action." Combined with Dreame's August 23 announcement of a 10-billion-yuan robotics industry VC fund with local governments, WRC 2026 was no longer just a trade show—it was the launch event of an industry inflection point.

The "97%" Number, Examined

The figure comes from a report jointly released by CAICT and the China Institute of Electronics at WRC 2026, and matches Counterpoint Research's concurrent estimate of ~22,000 global units shipped in H1, up nearly 300% year-over-year. By vendor:

AgiBot (Zhiyuan): ~9,700 units in H1
Unitree: over 7,000 units
Galbot (Yinhe General): over 1,100 units
Top five vendors combined: ~86% of the global market

Switching the scope to total global humanoid robot shipments, Chinese vendors reach 97%—the remaining 3% is scattered among Figure, Tesla Optimus, Agility Digit, and a few others. In other words, the "global humanoid robot" market has effectively been taken over by China.

165.5 Billion Yuan, 24.5% Growth

This is the MIIT-scope figure for the entire robotics industry (industrial robots + service robots + key components). For comparison: ~130 billion yuan in 2024, ~150 billion in 2025, and over 160 billion in just H1 2026—the full year will likely exceed 300 billion. This runs a full year ahead of MIIT's earlier projection that 2025 would break 200 billion.

The Overlooked Number: Where Robots Actually Work

The application breakdown is the most easily ignored figure:

Entertainment/performance: 33.6%
Data production & research: 27%
Service & guided tours: ~19%
Smart manufacturing: 12.8%
Warehouse & logistics: 4.9%

Less than 20% have actually entered smart manufacturing and warehousing. Counterpoint's data exposes an awkward fact: despite explosive shipments, most humanoids remain at the performance and research stage. WRC highlights included Unitree's H1 running at 12.65 m/s, Galbot's ET1 bipedal "holistic intelligence" demo, and UBTech replicating a customer's production line 1:1 at a grasping rate near 1,100 pieces/hour—but scaling from one line to "scale" still requires continuous replication across hundreds of factories.

Three Hidden Inflection Points

1. From "tech demos" to "capital + scenarios." On August 23, a 10-billion-yuan robotics VC fund was launched by Dreame with local governments, targeting core components, robot bodies, and scenario applications. It's not isolated: in August alone, Qingyan Technology closed a nine-figure-yuan Series A (led by Legend Capital), UBTech began building a 10,000-unit-capacity super factory with Siemens, and Galbot disclosed scaled operations in 10+ logistics centers.

2. Government enters the field. MIIT and SASAC jointly issued a 2026 humanoid robotics and embodied AI special program, explicitly requiring "100+ high-value scenarios" for real-world training—pulling the industry into a three-wheel drive of government procurement + central-SOE scenarios + state VC. The last industry to enjoy this was EVs, whose penetration went from 5% to 50%+ between 2020 and 2024 under a similar mechanism. This round's playbook is highly similar.

3. Data + training paradigm shift. At WRC, CloudTech released RoboForge, a system-level embodied brain co-developed with Nanyang Technological University, and proposed a new "on-the-job verification" evaluation system. UBTech's Thinker-WLA world model topped the LIBERO benchmark with 176% on-device inference efficiency improvement. Embodied AI is no longer just comparing "whose body runs faster" but "whose world model transfers better to unfamiliar scenarios"—an early signal of a fork between the "data is king" (scaled real data) and "model is king" (synthetic data + physics simulation) routes.

Staying Clear-Eyed

Tencent Tech noted on August 22 a counterintuitive phenomenon: audiences are suffering demo fatigue—"last year it grabbed a cup, this year it's still grabbing a cup." The marginal utility of performance demos is decaying fast. Unitree founder Wang Xingxing admitted at WRC: "Why haven't we scaled yet?" His judgment: the industry hits its true industrialization threshold when robots can enter arbitrary unfamiliar home environments and independently complete ~80% of tasks.

Conclusion

Putting the week's threads together—97% shipments, 165.5 billion yuan scale, 10-billion-yuan fund, the MIIT special action, the model-route fork, the new on-the-job verification standard—embodied intelligence has moved past the "single-point muscle-flexing" stage into a tug-of-war of "systems engineering + capital + policy + scenarios." Whoever pushes open the door of "80% of tasks in unfamiliar scenarios" first wins the next round. WRC 2026 is the starting point where that crack became visible—not the end.