Year One of Humanoid Robot Mass Production: China's Annual Output Expected to Exceed 100,000 Units

nashnova research
今天发布阅读约 14 分钟

China produced over 40,000 humanoid robots in H1 2026, with full-year output on track to exceed 100,000 units. Goldman Sachs raised its 2026 global shipment forecast to 75,000 — this means → the narrative is shifting from lab demos to factory deliveries.

01

What does 100,000 units actually mean?

China's Ministry of Industry and Information Technology (MIIT) reports H1 2026 output exceeded 40,000 units. Multiple brokerages now call this year China's humanoid-robot "mass-production year one."
Goldman Sachs raised its 2026 global shipment forecast from 25,000 to 75,000 in an August physical-AI report. It projects roughly 890,000 units by 2030 and 6.5 million by 2035, implying a market worth about $138.3 billion — nearly 5× its prior estimate.
Morgan Stanley raised China's 2026 shipment forecast from 28,000 to 50,000.
⚠ Goldman's figure is "global shipments"; MIIT's is "China domestic output." The two use different scopes. In plain terms = they are not measuring the same thing — do not add or directly compare them.
02

Five new national standards — what do they govern?

MIIT approved 274 industry standards on November 16, including 5 specific to humanoid robots: training-ground construction, training-data management, intelligence-capability grading, electric-drive integrated joints, and dexterous hands.
This means → the industry now has its first formal "capability grading framework" — a standardized way to measure how smart a robot actually is.
Zheshang Securities notes that grading results will feed into certification and procurement processes. Firms with strong vertical models and reproducible test records stand to enter government purchasing lists and pilot programs first.
Put simply = whoever passes the grading test and gets a scorecard first gets orders first.
03

What are the tech giants doing?

Tesla: Elon Musk says a live Optimus demo is coming within two months — unedited, no remote operation, no scripted task flows. The keyword is "authentic."
Samsung: Plans to unveil two in-house humanoid robots at CES in January 2027 — an entry-level HR and a premium HE with more joint degrees of freedom. This will be Samsung's first public showing of self-developed humanoids.
LG + Nvidia: Signed an August MOU. LG's robots will run on Nvidia's Isaac GR00T foundation model and Jetson Thor compute platform, with a public debut expected in Q1 2027.
CITIC Securities positions "physical AI" as the next wave of artificial intelligence, noting Apple, OpenAI, and Meta have all entered the space.
04

Where is the industry's center of gravity shifting?

Zheshang Securities argues that after hardware mass-production lands, the core bottleneck shifts from build-rate validation to embodied-AI intelligence capability.
This means → the next phase is no longer about "how many units can you build" but "how smart can the robot actually get."
High-quality interaction data is the essential input for embodied-model iteration. In plain terms = without data to feed the models, the flywheel stalls. The sector's main thread is extending from hardware into data and AI.
05

What are the Hong Kong-listed concept stocks doing?

UBTECH (09880): Signed a strategic partnership with China Telecom Terminal. Targets: home-companion robots past 1 million units, ultra-bionic and wheeled humanoids past 100,000 units. Frost & Sullivan ranks UBTECH #1 globally by revenue in both the humanoid-robot and full-size embodied-intelligence humanoid categories.
Horizon Robotics-W (09660): China Renaissance reiterated a "Buy" rating with a HK$5.80 target, citing rising market share in smart-driving chips — processors that turn road data into driving decisions — and accelerating domestic substitution of Nvidia.
Geek+-W (02590): Launched its "24/7 Autonomous Smart Warehouse" initiative, pushing toward fully unmanned 7×24-hour warehouse operations.
Dobot (02432): Debuted three new products at the China International Industry Fair. Its self-developed embodied large model, "Kongyi Dobot WAM," launched in May, breaks through spatial-perception bottlenecks in conventional vision-language-action models.
06

What is the next key test?

From MIIT output data to five new standards to concentrated moves by global tech giants, the "mass-production year one" narrative is fully laid out.
This reflects a core question the market is now digesting: can the production ramp actually support current valuations?
In plain terms = the story is rich, but the next real proof point is whether actual shipments and the embodied-AI data flywheel truly start spinning — that is this sector's next make-or-break moment.

市场有风险,内容仅供研究参考,不构成投资建议。