Head to head
US vs China.
Every key metric on this site, side by side — deployment, humanoids, supply chain, capital — plus the SemiAnalysis Unitree comparison in full.
Across 28 metrics:20 favor China4 favor the US4 contested
Industrial robot installations — US vs China vs rest of world
China installs more industrial robots each year than the rest of the world combined; US installations have been roughly flat for a decade. Dashed = IFR outlook. Cobots are counted inside the industrial total; humanoids and service robots are not (no country-level IFR split).
Deployment & density
| Metric | United States | China |
|---|---|---|
| Industrial robot installs (2024) | ~50K units across the Americas (~9% of world; US is most of it) | 295,000 units — 54% of all robots installed worldwide |
| Robot density (per 10K manufacturing workers) | 307 — 8th worldwide | ~470 — passed Germany and Japan in 2024 |
| Operational robot stock | Part of the Americas' installed base; no country near China's scale | Over 2 million robots in operation — the largest of any country |
| Humanoid installations (2025) | Most of the non-China remainder of ~16,000 units | >80% of all installations; AgiBot #1 at ~31% share |
| Single largest robot fleet | Amazon — 1M+ robots across 300+ fulfillment centers | Distributed across manufacturers; no single fleet at Amazon's scale |
Traditional arms vs cobots — the AI leading indicator
| Metric | United States | China |
|---|---|---|
| Traditional industrial arms
Neither superpower leads here: Japan + Europe (FANUC, Yaskawa, ABB, KUKA) hold ~85% of arm revenue. Between the US and China, only China produces at volume.
|
Not a producer — no US firm makes heavy-duty arms; Adept was sold to Omron, Rethink's assets to Germany's HAHN. Relies on Japanese & European imports | Secondary player — passed Japan in unit production share in 2024, but no Chinese firm tops 2% of the global arm market and median Chinese-firm margins run under 1% vs over 4% industry-wide |
| Collaborative robots (cobots) | A few small entrants (Standard Bots); negligible global share | Over half of global cobot shipments since 2023; after Denmark's Universal Robots (~30%), most of the top cobot makers are Chinese |
| Cobot self-sufficiency | Imports the bulk of its cobots | ~90% of domestic cobot demand met by Chinese makers — versus ~60% for industrial robots overall |
| Why cobots are the signal
Incumbents' high-precision, heavy-payload dominance matters less as inexpensive AI-integrated systems spread.
|
Leads the AI layer — foundation models and compute — but not the cheap, AI-integrated hardware that will carry it | Cobots are only ~12% of installs but proxy the AI-integrated future: cheaper, software-heavy, easy to deploy. China's cobot lead signals its likely position as arms go AI-native |
Humanoid race
| Metric | United States | China |
|---|---|---|
| Humanoid shipment leaders | Figure ~350 built (1/hour by spring 2026); Agility single-digit units per shift | Unitree 5,500 shipped in 2025 (20K target 2026); AgiBot 10,000 cumulative; UBTECH 1,000+ industrial units delivered |
| Entry humanoid price | No commercial sub-$50K product; Optimus targets <$20K at scale (unshipped) | Unitree G1 from ~$13.5–16K, commercially available today |
| Announced factory capacity
US nameplate rivals China's plans on paper; realized output is the gap
|
≈52K/yr design targets: BotQ 12K + RoboFab 10K + Hyundai's planned 30K Atlas plant — plus Tesla's 1M/yr Fremont ambition | Unitree 20K target for 2026 alone; UBTECH 10K/yr by 2027; ~200 humanoid makers |
| Labor-replacement economics | Pre-commercial — pilots at BMW, GXO, Mercedes | Teleoperated Unitrees already run at ~$23–26/hr, under the $30/hr human baseline; ~250 deployed in labor settings |
| Unit economics | All major humanoid OEMs pre-revenue or pre-commercial | Unitree: ~67% gross margin on G1; 2025 revenue ~$235M (+335%), profitable enough to IPO |
Supply chain & components
| Metric | United States | China |
|---|---|---|
| Humanoid component supply chain | Vertical integration islands (Tesla, Figure in-house actuators); near-zero merchant capacity | ~70% of global component supply; 63% of Morgan Stanley's Humanoid 100 suppliers are Chinese |
| Cost to build the same robot | A UR5e-class robot arm costs ~2.2× more to build in the US; an Optimus-class humanoid without Chinese parts: $131K vs $46K (~3×) | Baseline — components 20–40% cheaper, motors at 30–40% of Western cost |
| NdFeB magnet production | ~3–5K t/yr operating (first commercial output Dec 2025); ~25K t/yr announced by 2028 | ~300K t/yr installed — ~90% of world production; export-license regime since April 2025 |
| Rare-earth separation & refining | ~5% of global capacity; first US heavy-REE output in decades (Energy Fuels, Mar 2026) | ~91% of global capacity; ~99% of heavy rare earths |
| Precision reducers | ~0% domestic volume — assembly only (Harmonic Drive, Beverly MA) | ~40% and rising at 30–40% lower prices; Japan holds the high end |
| Lidar | Ouster the main merchant player ($169M revenue 2025); Luminar bankrupt Dec 2025 | ~89% of the global market (Hesai, RoboSense + 2 others); robotics lidar shipments up 1,459% YoY |
| Battery cells | Pack/BMS strength (Figure in-house); first US robot-grade gigafactory (EnerSys) starts 2028 | ~70% of global cell installations; CATL + BYD alone >55% |
| Dexterous hands | Research-grade, low volume (Psyonic; Shadow in UK) | Linkerbot claims >80% of high-DoF hands; 10,000th shipped Nov 2025 |
| Robot brains (compute & chips) | NVIDIA Jetson Thor is the de facto humanoid standard — used by Chinese OEMs too; Qualcomm, Ambarella alternatives | Dependent on US chip designs; domestic alternatives under export-control pressure |
| Robot foundation models | Physical Intelligence ($5.6B), Skild ($14B), NVIDIA GR00T, Figure Helix — the deepest bench | Fast followers; Unitree is the leading platform for humanoid AI research but the frontier models are American |
Capital & policy
| Metric | United States | China |
|---|---|---|
| Venture capital | $23B+ raised by robotics startups in 2026 YTD; Figure $39B, Skild $14B valuations | Humanoid firms raised ~$769M in 2023, ~$990M in H1 2024 — smaller checks, more state capital |
| Public-market access | Symbotic, Serve, Ouster listed; no humanoid IPO yet | Unitree's STAR Market IPO approved in a record 73 days at ~$6.2B; UBTECH already listed in HK |
| Government posture | DoD OSC loans + MP price floor (~$3B+ committed to magnets); robotics executive order pending; GUARD Act introduced | Embodied AI in the 15th Five-Year Plan; national humanoid standards (2026); 'Robot+' density-doubling roadmaps |
| Hardware iteration speed | Quarters — supply loops cross the Pacific | Weeks — proximate suppliers from the smartphone/EV ecosystem |
Case study: Unitree vs the U.S. humanoid stack
SemiAnalysis calls Unitree the proof of the 'DJI strategy' — bootstrap through researchers and hobbyists, own the bottleneck components, then ride ecosystem maturity into every adjacent market. Here is how it stacks against the American approach, dimension by dimension.
| Dimension | Unitree | US humanoid OEMs |
|---|---|---|
| Cumulative humanoids shipped | ~10,000th unit approaching (June 2026); 5,500 shipped in 2025 alone | Figure 350+ built; Agility single-digit volumes; Optimus not shipped externally |
| Price trajectory | $50K+ → $27.3K in 12–18 months; under $20K in some deals; G1 from ~$16K | No public price; BOM alone ~$46K with Chinese parts, $131K without |
| Gross margin | ~67% on the G1 — profitable while cutting prices | Pre-commercial; burn funded by venture capital |
| Vertical integration | Self-developed BLDC motors (30–40% of Western motor cost), planetary gearboxes, lidars, depth cameras | Figure builds actuators/packs in-house at BotQ; Tesla designs in-house but buys volume parts from Sanhua/Tuopu (China) |
| Decoupling status | G1 entirely decoupled from American components | No US humanoid is decoupled from China — magnets, reducers, screws all flow through Chinese chokepoints |
| Research-platform pull | The leading platform for humanoid AI research — NVIDIA's open GR00T reference humanoid is built on a Unitree H2 Plus chassis | US labs buy Unitrees to train American foundation models on Chinese hardware |
| Where the US still wins | Buys US chip designs; frontier autonomy models are not Chinese | NVIDIA compute, foundation models (Helix, π0, GR00T), capital depth, and Amazon-scale deployment demand |
Method: Side-by-side synthesis of every key US-vs-China metric tracked on this site, plus the SemiAnalysis Unitree and robotics-supply-chain analyses. The 'Traditional arms vs cobots' section (added 2026-06-18) draws on FAI's 'The State of Industrial Robotics' (Michael & Alden, 2026).. Captured as of Jun 12, 2026. "Edge" calls are editorial judgments on sourced data, not scores.