The race to put humanoid robots to work is accelerating fast. Venture capital poured $40.7 billion into robotics in 2025 alone , more than triple 2023 levels , and governments from Washington to Beijing are treating embodied AI as a matter of national priority. China has committed a staggering $138 billion state fund to AI and robotics. Pilots are running in factories and warehouses. Foundation models are being trained on real-world data.

But a new report from McKinsey & Company throws a bucket of cold water on the hype , not about whether humanoid robots will work, but about whether the industry can actually build enough of them. The culprit? The supply chain.

It’s Not the AI. It’s the Gearbox.

Most coverage of humanoid robots fixates on AI breakthroughs: better foundation models, smarter manipulation policies, improved locomotion. McKinsey argues that’s looking in the wrong place.

The real constraint is the hardware stack , and specifically, whether the suppliers who build the most critical components can scale fast enough to meet exploding demand.

A humanoid robot’s bill of materials breaks down into five main domains:

• Actuators , the motors, gearboxes, and drives that move the joints , make up 40–60% of total unit cost
• Sensing and perception systems account for 10–20%
• Compute and control platforms represent 10–15%
• Structural components cover 5–10%
• Battery modules take another 5–10%

Together these five domains account for 85–90% of what it costs to build a humanoid robot. The current price tag? Roughly $30,000 to $150,000 per unit. The widely cited target for mass-market viability is under $20,000 , meaning the industry still needs dramatic cost compression just to get started.

Where the Danger Zones Are

McKinsey maps the supply chain into three risk tiers.

Low Risk: Strong Industrial Foundations

Low risk components , brushless DC motors, camera modules, LiDAR, battery cells , already benefit from massive scale in adjacent industries. EV manufacturers like CATL, LG Energy Solution, Panasonic, and Samsung SDI have driven costs down for batteries. Camera and radar suppliers built for autonomous vehicles are transferable. These components still need adaptation for humanoid use, but the industrial muscle is there.

Medium Risk: Adaptation Required

Medium risk components , encoders, inertial measurement units, vision hardware , exist at scale but require more significant reworking. Autonomous driving hardware, for example, is often too bulky and power-hungry for a bipedal robot wearing it.

High Risk: The Real Choke Points

High risk is where things get serious. Three clusters stand out as potential bottlenecks:

1. Precision Motor Components

Harmonic and strain-wave drives , the compact, high-torque gearboxes used in humanoid joints , are produced by a tiny group of manufacturers including Harmonic Drive and Nabtesco, plus a handful of emerging Chinese players. These are precision-bound, capital-intensive parts that take a long time to qualify. Unlike electronics, you can’t just add a new fab line overnight.

Planetary roller screws may be the most acute risk of all. They serve a high-load, high-precision niche with a narrow supplier base, long lead times, and almost no substitution options. As humanoid robots push toward heavier payloads and more dynamic motion, demand could outrun supply capacity within the decade.

Permanent magnets add a geopolitical dimension. High-torque actuators depend on neodymium-iron-boron (NdFeB) rare-earth magnets , and China controls approximately 69% of global rare-earth mining and 90% of magnet processing capacity. Recent export licensing changes have already caused volatility. Elon Musk has publicly acknowledged that magnet supply issues have affected Optimus production.

2. Force and Tactile Sensing

Six-axis force/torque sensors, linear-force sensors, and tactile sensors are all underdeveloped. These components don’t benefit from automotive or consumer electronics spillover , they’re built by a handful of robotics-focused vendors and research-stage startups. Scaling them requires not just machining capacity but precision calibration infrastructure that barely exists at commercial scale.

3. Compute and Control

This is less a supply problem and more an integration problem. There’s no standardized, safety-certified “robot ECU” , the humanoid equivalent of a car’s engine control unit. Most OEMs cobble together Nvidia Jetson compute modules, automotive microcontrollers, and custom middleware. Tesla is the rare exception, having adapted its FSD compute stack for Optimus into a unified system , but few others have Tesla’s semiconductor capability.

The Scaling Dilemma

McKinsey identifies a classic chicken-and-egg problem at the heart of the humanoid supply chain. Suppliers won’t invest in dedicated production lines until volumes justify it. But volumes can’t grow until costs drop. And costs won’t drop without dedicated production lines.

Right now, most humanoid OEMs are forced to either vertically integrate , building components themselves , or engage in expensive one-off co-development deals with a small number of partners. Neither approach is cost-efficient at scale.

Vertical integration is rational given today’s conditions: it protects IP, controls cost, and reduces dependence on suppliers who may not deliver. Tesla builds its own frameless motors and gearboxes. Agility Robotics heavily modified off-the-shelf motors to make Digit’s legs work. But this approach can’t last forever as the industry tries to reach mass-market pricing.

The Land Grab Is Already Underway

Despite the premodular state of the industry, McKinsey notes that a wave of strategic partnerships over the past 18 months signals that major industrial players aren’t waiting around.

One large automotive and industrials supplier signed three humanoid actuator partnerships in five months , with Neura Robotics, UK-based Humanoid, and Chinese manufacturer Leju Robotics , positioning itself as a preferred actuator supplier across multiple platforms. Bosch entered through a partnership with Neura Robotics and formed a humanoid-focused joint venture in China. Magna International took an equity stake in Sanctuary AI.

On the compute side, Qualcomm launched the Dragonwing IQ10 , a humanoid-specific processor , and is working with Figure AI and Neura Robotics. Infineon, NXP Semiconductors, STMicroelectronics, and Texas Instruments have all formalized humanoid-specific product integrations.

In manufacturing, Jabil has become the worldwide production partner for Apptronik’s Apollo humanoid. SoftBank announced a $5.4 billion acquisition of ABB’s robotics division. Amazon has been quietly acquiring robotics companies including Fauna Robotics, Rightbot Technologies, and Rivr.

The message from McKinsey is clear: the companies that define standardized subsystem platforms now , actuator modules, compute stacks, sensor suites , will capture the ecosystem leverage that matters most when volumes finally scale.

What Suppliers Should Do Now

The report outlines five priorities for suppliers looking to stake their claim:

• Forge early co-development partnerships , embed components into robot platforms before specifications are locked
• Invest in safety and certification , SIL/ASIL certification and ISO compliance lower adoption barriers, especially in Western markets
• Design for modularity , interoperable components that serve multiple platforms will broaden addressable demand
• Prove scalable manufacturing , demonstrate volume readiness with automation, secured rare-earth supply, and quality control
• Build life cycle revenue streams , spare parts, maintenance contracts, and predictive diagnostics will be major profit pools as fleets grow

The Bottom Line

McKinsey’s verdict is that the humanoid supply chain is both the biggest constraint on the industry and, for those who move quickly, the biggest opportunity in a generation. The window to shape the supply chain is open , but narrowing fast.

One data point captures the stakes: a large automotive supplier went from zero humanoid presence to becoming a key actuator supplier in under two years. That’s how quickly incumbency can form once architectures stabilize and volumes start to scale.

The humanoid era is coming. Whether it arrives on schedule , and at what cost , will be decided not in AI labs, but in factories making gearboxes, magnets, and sensors that most people have never heard of.

Source: “Turning humanoid supply chain constraints into billion-dollar wins,” McKinsey & Company, April 2026