Are Chinese humanoids autonomous, or is it teleoperation?

Both, and the distinction matters when you buy. A lot of what ships as "autonomy" today is teleoperation plus scripted skills; real-world manipulation data remains scarce and expensive to collect, and that data bottleneck — not walking or balance — is the current gate on useful VLA-model autonomy. Ask vendors to separate what was demoed from what runs unattended in deployment, and to name which tasks are scripted.

Should I buy for research or for production deployment?

These are different purchases with different economics. Research and education units ship fast at published price floors (per GGII, basic hardware fell to ~¥100k, and Unitree lists R1 at ¥39,900 / G1 from ¥99,000). A production-grade unit fitted to a real factory process — integration, tooling, safety, operations, maintenance — lands closer to an all-in ¥500,000–600,000 (~$70–84k) per working station, which is why RaaS / per-station subscription models are gaining share for pilots.

How should I read vendors' unit and revenue numbers?

Separate delivered from ordered from intended — the three get blurred in marketing. Throughout this report, company-reported or single-source figures carry a † mark; treat those as unverified until you confirm them against listed-company disclosures or independent research. When a vendor cites a large "order," ask whether it is a signed contract, a letter of intent, or a stated target.

State of Chinese Humanoid Robotics 2026

Q: How mature is China's humanoid market in 2026?

It has crossed from demo to shipment — 2026 is widely called the first true mass-production year, with China holding an 80%+ share of global installations and IDC forecasting the domestic market roughly doubling to ~$1.3B. But maturity is uneven: the installed base still skews to research, education and showcase deployments, and factory work is dominated by pilots. Channel research (Goldman Sachs) puts the path from first meeting to stable deployment at roughly two years, so plan for validation cycles rather than off-the-shelf production capacity.

Q: What are the biggest import and compliance risks?

Two regulatory walls. US tariffs on Chinese robots currently run 25–125% by classification, with NDAA-2026-era restrictions further chilling government-adjacent procurement. In the EU, the AI Act's obligations begin binding key AI-system categories from August 2, 2026, layered on existing CE conformity and GDPR exposure for robots streaming video/audio/location data. Components and quadrupeds clear customs and support burdens more easily than full industrial humanoids, so the export mix skews that way.

Published July 2026 · Last verified: July 2026 · A Sourcebotics industry report

Executive summary. 2026 is the year Chinese humanoid robotics crossed from demo to shipment. China holds 80%+ of global humanoid installations, the domestic market is doubling to ~$1.3B (IDC), unit costs have fallen to ~¥100k for basic hardware, and exports grew ~210% YoY in Q1 — even as the US raises tariff walls and the EU AI Act adds compliance cost. The gating factor has shifted from "can it walk" to "can it work": deployment timelines, hands, and data remain the real bottlenecks.

1. Market size: doubling, and concentrated in China

Four load-bearing numbers frame the year:

Metric	2026 status	Source type
Global humanoid installations, China share (2025)	>80%	Third-party research
China humanoid market size, 2026	~$1.3B, >2x YoY	IDC forecast
Complete-robot makers / models on market	140+ / 330+	MIIT-affiliated tally
Registered humanoid-related companies	1,000+	Corporate registry data
Dexterous hand segment, 2026	$300M+, ~150% YoY	IDC

Adjacent context: commercial service robots add roughly $400M growing ~15%, and IDC notes ~60% of embodied-AI projects now deploy multiple form factors (humanoid + wheeled + quadruped) rather than betting on bipeds alone — a pattern buyers should copy.

2. The commercialization inflection — with honest caveats

2026 is widely called the industry's first true mass-production and commercialization year. The evidence: UBTech rolled its 1,000th Walker S2 off the line at end-2025 and delivered 500+ industrial humanoids in 2025†; AgiBot shipped 5,100+ units in 2025† targeting tens of thousands in 2026†; Galbot put 1,000+ wheeled units† into retail operation; EngineAI booked a 3,000+ unit patrol intent order†.

The caveats, which we consider load-bearing: today's installed base still skews to research, education and showcase uses; factory work is dominated by pilots. Channel research (Goldman Sachs) describes the standard adoption path as 3–6 month POCs run 2–3 times, ~12 months of validation, first orders under 50 units — roughly two years from first meeting to stable deployment. And much deployed "autonomy" remains teleoperation plus scripted skills, with data collection for VLA models the acknowledged bottleneck. Nobody serious disputes this privately; buyers should plan for it explicitly.

3. The cost curve: hardware deflates, deployment doesn't (yet)

Per GGII, the manufacturing cost of a basic humanoid unit fell to ~¥100,000 (~$14k) by Q1 2026 — a stunning figure against 2023's ¥500k+ baselines. Retail price floors followed: Unitree's R1 lists at ¥39,900, G1 from ¥99,000.

But deployment-grade economics move slower. A robot fitted to a real factory process — integration, tooling, safety, operations, maintenance — lands at an all-in ¥500,000–600,000 (~$70–84k) per working unit. The canonical payback math (¥500k robot vs. a ¥100k/year worker ≈ 5-year simple payback, halving under two shifts) explains both why high-wage overseas markets are priority targets and why RaaS / per-station subscription models are gaining share for pilots.

4. The players: a market with tiers, not a winner

Our full company guide covers profiles; the structural read is a three-tier market — scale deployers (UBTech, AgiBot), platform/ecosystem leaders (Unitree — including a 60–70% global quadruped share† and the deepest research install base; Fourier in healthcare), and a fast-reshuffling specialist tier (Robot Era in hands, EngineAI in patrol, Galbot in retail, LimX and others). Supply-chain leaders matter as much as robot brands: Realman shipped 100,000+ joint modules in 2025†, Inspire delivered ~2,000 dexterous hands†, and component vendors report 30–50% overseas revenue — the quiet headline of the year.

5. Globalization: exports are the fastest line in the industry

Q1 2026 humanoid-robot exports grew ~210% YoY, led by Europe, Southeast Asia and the Middle East. Concrete footprints: UBTech ships to 50+ countries with overseas+HK revenue ~¥475M (2025)†; AgiBot opened its first overseas experience center in Malaysia; Dobot's robotic coffee stations pour 45-second cups in the UAE; Geek+ runs warehouse deployments across Europe. Robot Era reports ~half its business overseas†. The export mix skews notably toward components and quadrupeds — categories that clear customs and support burdens far more easily than industrial humanoids, and (in our view) will out-grow full-humanoid exports through 2027.

6. Headwinds: two walls and two bottlenecks

Wall 1 — the US. Tariffs on Chinese robots currently run 25–125% by classification, with NDAA-2026-era restrictions further chilling government-adjacent procurement. The practical effect: US demand concentrates in research purchases and component flows, while vendors redirect commercial energy to GCC, SEA, Europe and LATAM.

Wall 2 — the EU. The EU AI Act's obligations begin binding key AI-system categories from August 2, 2026, layered on CE conformity and GDPR exposure for robots streaming video/audio/location data. Europe stays open — Q1 exports prove it — but compliance cost now favors vendors (and intermediaries) who can hand buyers a complete conformity story.

Bottleneck 1 — hands. Dexterity remains the hardest hardware problem; hence a $300M+ hand market growing 150% and humanoid makers vertically integrating hands.

Bottleneck 2 — data. Real-world manipulation data is scarce and expensive to collect; whoever industrializes collection (teleop farms, simulation, deployment fleets) sets the pace of useful autonomy.

7. Capital: the IPO wave

Unitree's listing process reached final stages in 2026†, headlining a queue of humanoid and supply-chain IPOs. Expect public-market discipline to do two things: force disclosure that separates delivered from ordered from intended units — a distinction this report enforces with † marks — and concentrate capital on the technical leaders, squeezing the long tail of 140+ makers.

8. Twelve-month outlook (our calls, on the record)

Entry research humanoids under $5,000 by mid-2027 as the R1 price floor is undercut.
Industrial orders concentrate in logistics handling, inspection and patrol — not general assembly.
RaaS passes one-third of new industrial pilots, shifting integration risk to vendors.
Components out-export whole humanoids (hands, modules, sensors) by revenue growth rate.
GCC + Southeast Asia become the #1 destination bloc for whole-robot exports as US friction persists.

We'll grade these publicly in the 2027 edition.

Methodology & sourcing

Figures aggregate IDC and GGII estimates, MIIT-affiliated tallies, listed-company disclosures, and manufacturer statements as of July 2026. Company-reported or single-source numbers are marked †. Currency at ¥7.15/USD. Corrections: hello@sourcebotics.com — we amend publicly.