Market and Technology Trends

Humanoid Robots 2026

Jan 2026

Chinese humanoid cluster: policy, pilot line, half of maker, faster iteration, and local supply at scale.

YINTR26535

Key Features

2025-2035 forecast in volume and revenue for humanoids
2025-2035 forecast in revenue for key semiconductors embedded (SoC, Memory, Vision 2D, Vision 3D, IMU, FT sensors, tactile sensors)
2025 expected market share of humanoids manufacturers
BOM estimation and semiconductor content

Report's objectives

To provide an in-depth view of the humanoid ecosystem from devices to end-users:

Mapping the ecosystem, from potential suppliers to manufacturers and end-users.
Overview of partnerships and pilots with large companies
Analyze the China-centered supply chain.

Deliver a quantitative outlook tailored to humanoid applications between 2025 and 2035:

Revenue and unit forecast
Adoption waves expected
BOM and cost structure
Semiconductor content

To provide crucial technical insights into and analyses of future technology trends and challenges:

Key technology choices
Technology dynamics
Safety and standards

Building useful humanoids: dexterous hands, cost compression, compliance

Humanoids are coming now because AI, hardware, and computing have converged. LBMs let robots learn and act with little coding, while LLMs enable natural interaction. Lighter, cheaper actuators and multi-hour batteries make them practical, and compact 200-TOPS computers give them real-time intelligence, finally making broad-skill humanoids viable.

The global humanoid robot market is set to increase from $0.6 billion in 2025 to $6 billion in 2030 at a 56% CAGR, then accelerate to $51 billion by 2035, at a 55% CAGR2030–2035. Based on a 6-year replacement cycle, our model predicts that ~8,000 robots will be sold in 2025, rising to ~136,000 in 2030 and 2.1 million in 2035.

The BOM is dominated by mechanics, dexterous hands, joint actuators, servomotors, and controllers, while silicon remains a smaller slice of the ASP. We estimate that semiconductors will constitute ~8% in 2025 and trend down toward ~5% by 2035 as compute and perception silicon benefit from scale and integration.

Adoption unfolds in three waves. First wave, Industrial (now): Pilots and early rollouts concentrate on intralogistics and light final-assembly support in brownfield plants and large warehouses, where ROI comes from ergonomic relief and labor gaps, rather than one-for-one station replacement (e.g., Apptronik Apollo at Mercedes-Benz for tote delivery, Agility Digit pilots for bulk handling).

Second wave, Consumer (next): price compression from Chinese OEMs (Unitree R1 ASP is ~$5.6k) expands education, developer, and prosumer experimentation, with constrained payload, durability, and safety versus industrial units.

Third wave, Medical (later): regulation and liability slow bedside roles, but policy signals, notably China’s State Council guidance to promote humanoids in elder care, pointing to earlier use in rehab support, mobility aids, and ward logistics before transitioning to direct patient handling.

In terms of investments, it means that humanoid manufacturers need to focus on dexterous hands to make it human-like and enable more opportunities for robust manipulation that will correspond to industrial applications. Once it is achieved, the goal will be to decrease the cost of humanoids to reach consumer applications. After that, safety and regulatory compliance will be necessary to unlock the third wave.

Funding, partnerships, and the Chinese ecosystem in humanoids

When looking at the humanoid robot manufacturers around the world, we listed more than 60 active companies. More than 50% of them are in China, and 12 companies are in the US. In Europe, there are relatively few companies, mainly startups.

Since 2017, cumulative funding has reached $9.8 billion (as of Oct. 25), with two companies, UBTECH and Figure AI, totaling $4.3 billion. These companies are leading in terms of funding but not in terms of shipments. Regarding expected shipments in 2025, Unitree is leading with almost twice the market share of Tesla, with Agibot completing the podium.

On the demand side, leading companies are partnering with large companies involved in automotive manufacturing or logistics tasks for pilot programs. Multiple demos have been shown from Apptronik, Figure AI, Agility, Astribot, and UBTECH.

Beyond industrial programs, Unitree is a strong player in R&D, education, and seed integrators. The latest Unitree R1, launched at $5.6k, is expanding access to humanoids for more creators and developers.

Regarding the ecosystem, China is pushing strongly in this direction. China’s MIIT has set a plan for 2023–2025 to establish an end-to-end humanoid innovation environment and secure core components. Of course, Nvidia remains a strong leader in the computing aspect, but local players such as Horizon Robotics and Cambricon, amongst others, are increasingly involved, following similar trends in automotive computing. For other parts like sensors, actuators, batteries, and encoders, the Chinese ecosystem is already quite strong and capable of supplying parts in large quantities.

Where performance comes from: hands, eyes, and rules

Humanoid performance is being pulled by three interacting stacks: mechanics, perception, and safety/robustness.

On the mechanical side, the BOM’s core remains in actuators and dexterous hands, the parts that unlock real factory work (bin moves, kitting, light fastening) and drive most of today’s engineering effort. Dexterous hands represent, on average, 31% of the BOM, and actuators around 42%. While actuators are quite mature and are already found in industrial robot arms, dexterous hands are becoming a strong focus area in engineering development.

On the perception side, the core is camera-dense vision with optional depth, feeding an increasingly capable on-robot AI. NVIDIA’s roadmap, including the Isaac + GR00T foundation models and Jetson Thor, is becoming a common standard for multi-camera pipelines and imitation/LLM-driven skills, which explains why head and torso camera counts and model sizes are rising across OEMs.

At the sensor level, we see mature choices for manipulation and state estimation. Cameras and multiple IMUs are commonly used by humanoid manufacturers. Depth sensing can be added for more precision and safety, but various concepts co-exist, such as using a capacitive skin as a safety feature.

For manipulation, 6-axis force torque sensors are used in each robot as well as inductive and magnetic encoders. The difference between entry-level and industrial robots is related to the performance resolution of these parts. To enable precise control and human-like manipulation, high-resolution sensors and encoders are needed in addition to tactile sensors in the hands. This strongly impacts the selling price of humanoids and explains why the ASP of some humanoids is more than $100k.

On the safety side, norms and standards are being developed, also pushed by Chinese organizations such as the China Institute of Electronics.