The 61% Dependency Paradox: Why the US–China Humanoid Robot Race Is Not a Zero-Sum Game

The global humanoid robotics industry is often framed as a binary US–China race: America owns the "brain" (AI chips, foundation models, simulation platforms) and China owns the "body" (manufacturing scale, component supply chains, deployment velocity). The reality in mid-2026 is far more nuanced — and far more interdependent.

US humanoid OEMs source 61% of their core hardware components from Chinese suppliers, according to a Wall Street Journal supply-chain audit of ten American humanoid startups. Meanwhile, Chinese firms still depend on US-designed AI accelerators for 87% of their training compute and on Japanese-made RV reducers for 73% of high-torque joints. Neither side can decouple without sacrificing at least two years of progress.

This article examines the five-layer interdependence stack, how the "GUARD Act era" of export controls is reshaping — but not severing — these ties, and which companies are positioned to capture value in a world of "managed interdependence" rather than full decoupling.

1. The Five-Layer Interdependence Stack

The humanoid robot supply chain is not a single pipeline. It is a five-layer stack, and the US and China dominate different layers — sometimes in complementary ways, sometimes in contested ones.

Layer 1: AI Compute & Foundation Models — US Dominance

At the top of the stack sits the "brain": AI chips, foundation models, and simulation platforms.

• NVIDIA commands an estimated 92% share of the AI training accelerator market used by humanoid robotics companies, according to TrendForce's Component Supply Chain Influence Index (CSCII). Its GR00T platform and Omniverse/Isaac simulation suite are the de facto standard for Western OEMs.
• Qualcomm leads in edge inference chips for robotic platforms, with its Robotics RB3/RB5 platforms powering a majority of Western mobile robots.
• Foundation model providers — OpenAI, Google DeepMind, Anthropic — are all US-headquartered and supply the large-model backbones that humanoid robotics companies fine-tune for embodied tasks.

Chinese humanoid firms, despite domestic alternatives like Huawei Ascend and Horizon Robotics, still rely on NVIDIA hardware for 87% of their pre-training compute, according to Alpine Macro's June 2026 supply chain report. Domestic chips are improving rapidly but remain 2–3 generations behind in energy efficiency for large transformer models.

The dependency metric: China → US AI compute dependency = 87%

Layer 2: Sensors & Perception — Split Market

The sensing layer — cameras, LiDAR, IMUs, force-torque sensors — is the most geographically distributed.

• Cameras and vision modules: Chinese suppliers (Huawei, HiKvision, Sunni) dominate volume production, capturing 68% of global unit shipments. Western OEMs including Figure AI and Tesla use Chinese camera modules in their non-sensitive models.
• LiDAR: Chinese firms (Hesai, RoboSense, Leishen) hold 70%+ of the automotive-grade LiDAR market, which has become the default source for humanoid robot perception systems due to cost and maturity advantages.
• Force-torque sensors: European and Japanese suppliers (ATI Industrial Automation, OnRobot, Keyence) still lead in high-precision six-axis force sensors, though Chinese firms like Vigatec are closing the gap, now reaching ±0.5% full-scale accuracy at 35% of the import price.

The dependency metric: US → China vision sensors dependency = 68% of units

Layer 3: Actuators & Motion Control — The Contested Middle

Actuators are the most competitive layer, with both sides racing for dominance. This is also where the most economically significant component trade flows occur.

• Harmonic drives: Japanese company Harmonic Drive Systems still holds ~55% global market share in precision harmonic reducers, with Sumitomo and Nabtesco commanding the RV reducer space. Chinese suppliers (Leaderdrive, Hengfengtai, Qinchuan) have grown rapidly, now capturing 38% of the global harmonic drive market in unit terms, but their high-end (≤1 arc-min backlash) yield remains 20–30 percentage points below Japanese peers.
• Servo motors and motor controllers: Chinese suppliers dominate low-to-mid-range servos, but high-torque-density integrated joint modules above 150 Nm still rely on Japanese and German technology.
• Planetary roller screws: European firms (Ewellix, Rollvis, SKF) lead this critical linear actuation technology for humanoid knee and hip joints. Chinese alternatives exist but struggle with lifespan above 10,000 km of walking equivalent.

American humanoid OEMs source roughly 55% of their actuator components — motor cores, position sensors, custom machined housings — from Chinese suppliers, the WSJ audit found. "We tried three US domestic suppliers for a custom joint module," one engineer told the Journal. "Either the precision didn't meet the motion spec, or the unit cost was 4× higher. We ended up going with a Shenzhen supplier."

The dependency metric: US → China actuator components dependency = 55%

Layer 4: Batteries & Power Systems — Chinese Near-Monopoly

The battery layer is where Chinese dominance is most complete — and most consequential for humanoid robot economics.

• Lithium-ion cells: Chinese manufacturers (CATL, BYD, EVE Energy, Ganfeng) produce ~73% of global lithium-ion battery cells and an even higher share of custom-pack solutions for robotics applications.
• Battery management systems (BMS): Chinese BMS chips and modules dominate the mid-range market, while Texas Instruments and Analog Devices still lead in ultra-high-reliability industrial BMS.
• Structural battery packs: The trend toward structural, load-bearing battery packs in humanoid robots — pioneered by Tesla and rapidly copied by Chinese OEMs — relies on Chinese cell supply and pack integration expertise.

For humanoid robots, where battery weight and energy density directly determine runtime (the "90-minute wall" we analyzed in our Battery Bottleneck feature), access to the latest cell chemistry is a competitive necessity. Most American humanoid startups source 100% of their battery cells from Chinese suppliers, simply because no domestic alternative can match the combination of energy density, cycle life, and cost.

The dependency metric: US → China battery supply dependency = ~100% for cell-level supply

Layer 5: Assembly & System Integration — China's Factory Floor

The final layer — mass assembly, quality control at scale, and rapid iteration manufacturing — is where China's advantage is most structural and hardest to replicate.

• "3 days prototyping, 7 days mass production": The Shenzhen–Dongguan–Suzhou manufacturing corridor has built an ecosystem where a new humanoid joint design can go from CAD file to a batch of 500 test units in under two weeks. In the US, the same process typically takes 3–6 months.
• Component localization rate: Chinese humanoid OEMs report 70–85% component localization, meaning the vast majority of parts come from within a 100-kilometer radius. For US OEMs, the figure is closer to 20–30% for core components.
• Cost structure: This manufacturing density translates directly into price. Unitree's G1 robot at ¥99,000 (~$13,700) undercuts comparable Western platforms by a factor of 4–6× — and the gap is structural, not temporary.

TrendForce's CSCII rates China as "Dominant" in the Power Plane (batteries, power electronics) and "Strong" in both the Mental Plane (AI, though still behind the US) and Movement Plane (actuators, structural components). The US is rated "Dominant" only in the Mental Plane (AI chips, foundation models).

2. The Export Control Architecture: A Leaky Wall

Against this backdrop of deep interdependence, both governments are building regulatory walls — but the walls are full of holes, and both sides are actively finding ways around them.

The US Side: GUARD Act and Beyond

The Generative AI United Responsibility and Disclosure (GUARD) Act, introduced in the US Senate on June 3, 2026, represents the most ambitious piece of AI regulation in US history. For humanoid robotics, its key provisions include:

• Export controls on "frontier AI models" trained on more than 10²⁵ FLOPs, requiring government approval before providing access to entities in "countries of concern" (which includes China).
• Chip export restrictions extending the October 2022 and 2024 rules, further tightening the threshold for AI accelerators that can be sold to Chinese entities. The latest rule, effective July 2026, lowers the performance density threshold from 4,800 to 3,000 TOPS·mm⁻².
• Supply chain disclosure requirements for federal contractors using humanoid or autonomous robots, mandating disclosure of "critical components" sourced from "foreign adversaries."

However, enforcement is uneven and loopholes abound:

1. The "robot chip loophole": Many edge inference chips used in humanoid robots fall below the current performance thresholds, meaning they can still be exported. Horizon Robotics and other Chinese chip designers have deliberately optimized their products to stay just under the threshold.
2. Third-country transshipment: Chips flow through Vietnam, Malaysia, and Mexico, often with minimal value-add, before reaching Chinese customers. US export control enforcement agencies are understaffed to police this at scale.
3. Open-source workaround: Chinese labs increasingly use open-source models (Llama, Qwen, DeepSeek) trained on domestic compute clusters, bypassing the need for US model API access entirely.

The China Side: Standards, Subsidies, and Import Substitution

China's response has been three-pronged:

1. National standards: The HEIS 2026 (Humanoid Embodied Intelligence Standard) framework, released in January 2026, establishes China's first comprehensive national standard for humanoid robot safety, performance, and interoperability. By setting its own standards early, China aims to shape the global playing field — much as it did with 5G and EV battery standards.
2. Massive subsidy programs: The "Robot+" action plan, provincial-level manufacturing subsidies, and dedicated humanoid robot funds at both the central and local levels are channeling an estimated ¥50–80 billion ($6.9–11 billion) annually into the sector. Shenzhen alone has earmarked ¥10 billion for humanoid robot industrial development through 2028.
3. Import substitution campaigns: Domestic substitution mandates for government and SOE procurement are forcing rapid adoption of Chinese-made components, even when performance lags behind imports. This creates a captive market that funds domestic R&D — a classic infant-industry strategy.

The results are visible in the data: Chinese harmonic drive suppliers grew their global unit market share from 22% to 38% between 2024 and mid-2026, largely on the back of domestic demand. In AI chips, Huawei's Ascend 910C now powers an estimated 23% of Chinese large-model training clusters, up from 8% in 2024.

3. The European Wildcard: A Third Pole Emerges

The US–China binary narrative obscures a critical development: Europe is quietly building a third pole in the humanoid robotics supply chain, with particular strength in precision manufacturing, safety certification, and industrial integration.

Europe's Supply Chain Position

TrendForce's CSCII rates Europe as "Strong" in both the Sensing Plane and Movement Plane, with particular depth in:

• Encoders and precision measurement: German suppliers (Heidenhain, Sick, Leuze) hold a distinct edge in high-resolution encoders and industrial sensors — critical components for precise robot motion.
• Industrial integration and safety: German and Swiss automation firms (Festo, Schaeffler, Bosch, Siemens) bring decades of industrial automation expertise and are natural partners for humanoid robot deployment in European manufacturing.
• Regulatory leadership: The EU AI Act, taking effect in August 2026, positions Europe as the global standard-setter for AI and robotics regulation — a position of soft power that translates into market access requirements for all vendors.

The SKL Robotics / Humanoid Moment

The July 2026 announcement that SKL Robotics (branded "Humanoid") raised $152 million at a $1.35 billion valuation — becoming Europe's first pure-play humanoid robot unicorn — signals that Europe is no longer just a component supplier but a potential full-stack competitor.

Founded in 2024 and based in London, Humanoid has already signed binding deployment agreements with Schaeffler for 1,000–2,000 wheeled humanoid robots by 2032, and named Bosch as its contract manufacturing partner. Its KinetIQ "shared brain" platform for multi-robot coordination directly targets the industrial fleet management market.

This matters for the US–China dynamic because:

• European manufacturers now have a non-US, non-China alternative for humanoid robot deployment
• Europe's deep industrial base means it can scale faster than pure-play startups without manufacturing heritage
• The EU AI Act creates a regulatory "moat" that favors EU-based companies that can certify faster

4. Deployment Race: Volume vs. Value

The most visible metric of US–China competition — deployment volume — tells a dramatic story, but one that is more nuanced than the raw numbers suggest.

The Volume Story: China's Scale Advantage

The deployment numbers are stark:

• 2025 global humanoid shipments: ~13,000 units, with Chinese firms accounting for 84%+ (Omdia data, cited by 36Kr)
• Unitree alone shipped 5,500+ humanoids in 2025, targeting 10,000–20,000 units for 2026
• AgiBot deployed 10,000+ robots across various form factors as of mid-2026, claiming the #1 global position for 2025 with 5,100 humanoid shipments and 39% market share
• China's Q1 2026 robot exports reached ¥113.2 billion ($15.7 billion), with humanoid robot exports surging 210% year-over-year

China's Q1 2026 exports of humanoid and industrial robots grew so fast that the customs administration created a new dedicated tariff code ("intelligent bionic robots") to properly track the category — a bureaucratic signal that the product category has reached economic significance.

The Value Story: US Premium Position

But volume is not the whole story. When measured by revenue per unit and gross margin, American companies retain a significant lead:

• Figure AI's Figure 02, deployed at BMW Spartanburg, commands an estimated $250,000–$350,000 per unit at current low volumes, including integration and software services
• Agility Robotics' Digit has $300 million in contracted orders at price points estimated at $120,000–$180,000 per unit, with RaaS pricing of $10–12 per hour
• Tesla's Optimus, while not yet commercially available, is positioned for high-margin software subscription revenue on top of hardware

For perspective: if Unitree ships 15,000 units at an average of $15,000 each, that's $225 million in revenue. If Figure ships 500 units at $250,000 each, that's $125 million — with far higher margins and a software revenue tail.

The question for the next 2–3 years is whether China's volume advantage translates into a data advantage that closes the AI-capability gap — the "deployment loop" thesis. If more robots in more real environments generate more data that feeds better models, the gap could narrow faster than Western observers expect.

TrendForce explicitly raises this concern: "China is rapidly accumulating real-world data through large-scale deployment, extending its supply chain advantage into a data advantage. This could help China's humanoid robot industry form a positive cycle of 'deploy, collect data, update the model, redeploy.'"

5. Scenarios: Three Futures for 2028–2030

We see three plausible scenarios for how the US–China humanoid robot competition evolves through the end of the decade, with very different implications for companies, investors, and policymakers.

Scenario A: Managed Interdependence (55% probability)

The most likely outcome is neither full decoupling nor full convergence, but a "managed interdependence" regime in which:

• Both sides maintain targeted export controls in sensitive areas (frontier AI training, high-end military-grade components)
• Commercial trade continues in most civilian and industrial applications, with increased scrutiny and compliance costs
• Supply chains partially "friend-shore," creating parallel ecosystems (US-allied supply chain vs. China-centered supply chain) but with significant overlap at the component level
• Europe becomes the swing market, courted by both sides and able to set regulatory standards that both must meet

In this scenario, companies that operate across both ecosystems — with products certified for both Western and Chinese markets, and supply chains diversified enough to withstand shocks — capture the most value.

Scenario B: Accelerated Decoupling (25% probability)

If geopolitical tensions escalate significantly — around a Taiwan contingency, new sanctions episodes, or major AI safety incidents traced to Chinese-origin systems — the walls could go up much faster:

• Broader export controls covering mid-range AI chips and more component categories
• Reciprocal Chinese restrictions on rare earths, battery materials, and critical components
• Forced bifurcation of standards, software platforms, and even basic protocols
• A genuine "two worlds" scenario with separate technology stacks

In this scenario, the biggest losers are globally oriented OEMs that built their business models on cross-Pacific supply chains. The biggest winners are vertically integrated players on both sides — Tesla/Figure in the US, BYD/AgiBot in China — that can absorb the cost of building redundant supply chains.

Scenario C: Convergence and Mutual Recognition (20% probability)

The least likely but highest-impact scenario is a gradual convergence toward a global humanoid robot market with shared standards and open supply chains. This would require:

• Bilateral or multilateral agreements on robot safety, data governance, and export controls
• International standards bodies (ISO, IEC) reasserting primacy over national standards
• The economics of scale simply overwhelming protectionist pressures

In this scenario, the entire market grows faster, and the winners are the companies with the best technology and execution regardless of nationality.

6. Investment Implications and Company Positions

The current state of interdependence creates specific investment opportunities and risks. Here's how we see the positioning of major players as of mid-2026:

Well-Positioned Companies

Companies Facing Structural Pressure

7. Conclusion: The Real Race Is Not Binary

The US–China humanoid robot race is often presented as a winner-takes-all competition. The evidence from mid-2026 suggests otherwise.

The reality is a layered, interdependent system in which each side has structural advantages that are not easily replicable by the other. The US cannot quickly rebuild a precision manufacturing ecosystem at scale; China cannot quickly catch up in frontier AI model capability. Both sides are trying — and both will make progress — but the timeline is measured in years, not quarters.

The companies that will win this decade are not the ones that pick one side or the other. They are the ones that navigate the interdependence skillfully: understanding where dependency is safe, where it is risky, and how to build resilience without sacrificing speed.

As TrendForce puts it: "Over the long run, what will determine the global humanoid robot industry's landscape is whether companies can build business models that combine scale deployment, data accumulation, and continuous model improvement." That combination requires both hardware scale and AI sophistication — and right now, no single country has both at the frontier.

The real race is not US vs. China. It's a race to see which ecosystem can integrate the global supply chain most effectively — while navigating the political pressures that keep trying to pull it apart.