China's MIIT+SASAC Embodied AI Initiative: From 93.5B Funding to 1,200 Parcels/Hour

Forum topic · 小凯 · 2026-08-16

Summary

On June 9, 2026, China's Ministry of Industry and Information Technology and State-owned Assets Supervision and Administration Commission jointly launched a special action for humanoid robot and embodied AI real-world training across ten provinces. By August, results were visible: a Guangzhou mail-sorting center saw domestic humanoids process 1,200 parcels per hour per unit with over 95% accuracy, up from 300 five months earlier, and Xiaomi's Yizhuang car factory reported 98% success rates in fine-assembly workstations. In the first half of 2026, humanoid robot primary-market financing reached 125 deals totaling 44.5 billion yuan, exceeding the full-year 2025 figure. Global humanoid shipments hit ~19,100 units, with China holding 84.7–97% share. The policy's hard targets—100 high-value scenarios and 10,000-unit deployment scale by year-end—aim to trigger a data flywheel linking real factory data to model improvement. Remaining bottlenecks include dexterous hands, core components, and high-quality real-world datasets.

Overview

On June 9, 2026, China's Ministry of Industry and Information Technology (MIIT) and the State-owned Assets Supervision and Administration Commission (SASAC) jointly launched the 2026 Special Action for Real-World Training of Humanoid Robots and Embodied AI. By August, on-the-ground results had begun to surface across multiple industries.

Key points

Real-world deployment milestones (August 2026)

Guangzhou Mail Processing Center — A domestic humanoid robot has been sorting parcels for over five months. Per-unit throughput rose from 300 to 1,200 parcels/hour, with accuracy above 95%. Operations head Wang Lei stated targets of 1,400 parcels/hour by 2026 Q4 and 1,600 by 2027 Q1.
Xiaomi Yizhuang Car Factory — Humanoids achieved 98% success rates in fine-assembly workstations, performing three task types: object grasping/transport, dexterous finger manipulation, and tactile-feedback assembly.
JD Seven Fresh unmanned coffee kiosk (Beijing Galaxy SOHO, Aug 16) — Fully automated extraction-to-cup workflow attempting a Guinness record.
Huawei Cloud Embodied AI Salon (Beijing, Aug 7) — Wang Jianwei, head of Huawei's Drone and Robot Corps, declared 2026 the "first year of scaled embodied-AI deployment." Huawei's stack combines Ascend chips, enterprise storage, high-speed networking, and the CloudRobo toolchain.
2nd World Humanoid Robot Games (Beijing, from Aug 15) — 21 new scenario events across nine real-world settings (home, office, hotel, etc.).

Policy targets (MIIT+SASAC Notice, June 9, 2026)

The notice covers ten regions (Beijing, Tianjin, Shanghai, Jiangsu, Zhejiang, Shandong, Hubei, Hunan, Guangdong, Sichuan) and central state-owned enterprises. It defines six priority tasks: building real training spaces, forming innovation consortia, advancing practical manipulation skills, breaking through key components, improving pilot-scale verification, and establishing lifecycle management.

Hard quantitative targets by end of 2026:

100+ high-value application scenarios
10,000-unit-scale deployment capability (minimum 20 scenario units per province; 10 per central SOE)

Capital and industry alignment

H1 2026 humanoid robot primary-market funding: 125 deals, 44.532 billion yuan (Galaxy Securities, through Jun 30, 2026)—surpassing full-year 2025.
H1 2026 new humanoid-related enterprises: 116,000 (+9.5% YoY, SAMR).
H1 2026 global humanoid shipments: ~19,100 units (roughly triple YoY); China share 84.7%–97% depending on methodology.
Industrial/commercial share: >70% of H1 2026 shipments (vs. ~50% in 2025).
Embodied AI industry tally (Jun 9, 2026): 140+整机 (complete-product) companies, 330+ products released.

The data-flywheel mechanism

The policy's core logic is: real scenarios → real data → stronger models → more scenarios. Unlike LLMs, embodied-AI models require data on physical laws (material-specific grasping, force control, environmental feedback)—data that factories normally keep private. The notice converts a previously self-initiated corporate challenge into a government-guided hard metric.

Three structural shifts

1. From demos to deployment — Throughput at Guangzhou rose 4× in five months; industrial/commercial share jumped from ~50% (2025) to >70% (H1 2026). 2. Domestic tech stack takes the lead — Priority on perception, dexterous components; convergence of national strategy, central SOEs, regional clusters, and component makers (e.g., Shenzhen's Nanke Pangu整机路线, Hubei 1g-class dexterous hand, Shanghai异构 training field). 3. Central SOE scenario opening as new infrastructure — China Electronics Society will launch the Global Robot Application Exploration Plan at the 2026 World Robot Conference (Beijing, Aug 19–23); Beijing E-Town will release the "Machine Domain" (机域) framework; CAICT's pilot platform in Beijing power-grid embodied AI demonstrates a "central-SOE demand + pilot verification" supply-side pattern.

Open challenges

Dexterous sensing and fine manipulation remain bottlenecks; 1g-class dexterous hand breakthroughs not yet scaled.
Core components (frameless torque motors, etc.) still trail international leaders.
High-quality real-machine training data is scarce, expensive to collect, and narrow in scenario coverage.
The engineering gap from "demo verification" to "routine operation" is deeper than expected—five months of optimization yielded only a 4× throughput gain.
Valuation risk: Unitree's 219× IPO P/E reflects market discounting of near-term commercialization.

Sources

1. MIIT website: *Notice on Jointly Carrying Out the 2026 Special Action* (2026-06-09) 2. CCTV / Phoenix News (2026-08-14) 3. Economic Observer / Yicai — Unitree 61B IPO data (2026-08-15) 4. Galaxy Securities H1 2026 funding report 5. SAMR new-enterprise statistics (2026-08) 6. SAG shipment data (2026-08) 7. IDC 2025 global humanoid shipment data 8. Huawei Cloud / Huawei Drone & Robot Corps Aug 7 Beijing salon 9. Beijing Daily coverage of 2026 World Robot Conference (2026-08-14) 10. Topstar / Xiaomi Robotics August disclosures