Actuators are up to 60% of a humanoid’s cost, and the US is short of the gears inside them

Apptronik chief executive Jeff Cardenas said on Fox Business on September 22, 2026 that the US has “major shortages in things like gears where we don’t have the supply base that we need.” Actuators, the motor-gearbox-sensor units that move each joint, can account for up to 60% of a humanoid’s bill of materials, he said, and Apptronik has designed more than 80 distinct electric actuators since it was founded.

Why it matters: the most expensive part of a humanoid is also the one with the thinnest supply base outside China, Japan and Germany. Whoever controls actuator supply in 2027 controls how many robots actually ship.

Why is there a shortage of gears for humanoid robots?

Because the precision gears a humanoid joint needs are made by a small number of specialists, and few of them are in the US. Cardenas did not name the missing parts beyond gears, and he gave no volume figures. His point was about the supply base: “It’s not just enough to have the software. We’ve also got to have world-class hardware to go along with it.”

The gear in question is usually a strain wave or planetary reduction stage. It turns a fast, low-torque electric motor into a slow, high-torque joint, and it has to do so with almost no backlash. A humanoid has dozens of them. Making them is a machining and heat-treatment business in which Japanese and German suppliers have long led, and in which Chinese suppliers have added capacity quickly. TrendForce’s July map of the sector, which we covered in who controls the humanoid supply chain, put China broadest across components, the US ahead in AI compute, and Japan and Germany strongest in mechanics. Cardenas’s comment is the American robot maker’s version of the same finding.

A shortage here does not mean robots cannot be built. It means US makers pay more, wait longer or import from the regions their policymakers most want to reduce reliance on.

How much of a humanoid’s cost is the actuator?

Up to 60% of the bill of materials, according to Cardenas. That is the high end of the estimates we have reported. Schaeffler, when it presented its formed strain wave gearbox in August, put actuators at around half of a humanoid’s build cost. The difference depends on how many joints a robot has, how dexterous its hands are and how much compute it carries.

Either figure makes the actuator the line item that decides whether humanoid prices fall. Schaeffler’s own answer is to form the strain wave gear rather than machine it, which it says cuts gearbox cost by 25% and material use by 75%, with mass production planned for 2027; the detail is in our report on Schaeffler’s formed gearbox. The hands add a separate layer of small actuators, compared in our robot hands guide.

Who makes humanoid actuators in-house and who buys them?

The industry is split, and the split is now visible. Some robot makers design and build their own joints to control cost and performance. Others sign long-term supply contracts with component makers and put their engineering effort into software and integration. The table below lists only arrangements that the companies or their filings have made public.

Humanoid actuator sourcing, as of September 28, 2026 (company statements, filings and reported deals):

Company	Robot	Approach	What is on record
Apptronik (US)	Apollo	Designs in-house	More than 80 distinct electric actuators developed since founding (CEO, September 2026)
Unitree Robotics (China)	G1 and other models	Builds in-house	Motors, reducers, joint modules and servo drives self-developed, per its IPO prospectus
Boston Dynamics / Hyundai Mobis (US/Korea)	Atlas	In-house within the group	Hyundai Mobis showed its first two humanoid actuators on September 10, 2026 as the basis for Atlas parts
ROBOTIS (Korea)	AI Sapiens	Builds in-house and sells to others	Supplies actuators to other humanoid makers; plans 1 million actuators a year of capacity in 2027
Humanoid (UK)	Company’s humanoids	Buys	Schaeffler to supply more than 50% of its joint actuators under a five-year deal (May 2026)
XPeng (China)	IRON	Buys from designated suppliers	Designation agreements with joint-module, actuator and precision-part makers signed September 22, 2026

The pattern is not East versus West. Unitree and Apptronik both build their own, as does the Hyundai group through Mobis, which we covered in Hyundai Mobis’s first humanoid actuators. XPeng and Humanoid buy, but with long agreements that tie suppliers to their production plans. ROBOTIS sits on both sides, as a robot maker that also sells joints. The makers that build in-house carry more capital cost and engineering risk. The makers that buy carry more exposure to exactly the shortage Cardenas described.

What did XPeng sign with its humanoid suppliers?

XPeng signed supplier designation agreements with makers of robot joint modules, actuators and precision structural parts at its first robotics supply chain partner conference, held in Guangzhou on September 22, 2026. It did not name the suppliers, say how many signed, or disclose contract values or volumes. The company said its production line is operational, that mass production is scheduled for the end of 2026, that the first commercial rollouts will be in XPeng stores and campuses, and that market launch and deliveries in China and overseas are planned for 2027.

Chief executive He Xiaopeng said humanoid robotics is more complex than the automotive industry and needs longer-term investment, according to Humanoids Daily. Five days later, Korea’s Robot Newspaper (iRobotNews) reported his remarks at a Guangdong AI conference that mass production starts in December, with deliveries in China from the second quarter of 2027 and overseas from the fourth quarter. XPeng commissioned the line itself on September 8, with more than 80% of core processes automated; see our report on the IRON line.

The supplier conference is the less visible step, and arguably the more important one. A car maker’s advantage in humanoids is not the factory floor. It is a purchasing organisation that already knows how to lock in parts suppliers years ahead. XPeng is applying that playbook to joints.

What does the actuator squeeze mean for 2027 production targets?

It makes actuator contracts a better guide to real output than announced unit targets. A company that has signed or built capacity for tens of thousands of joints can plausibly build hundreds or low thousands of robots. A company that has not has a target, not a plan.

By that test, the most concrete positions today belong to companies with named arrangements: Schaeffler’s five-year volume commitment to Humanoid, Hyundai Mobis’s in-group supply to Atlas, Unitree’s vertically integrated lines and XPeng’s newly designated suppliers. Apptronik’s 80-plus actuator designs show deep in-house expertise; its chief executive’s complaint shows that design is not the same as supply. US policy debate about humanoids has so far focused on AI models and on imports of finished robots. The gear shortage suggests the harder dependency is one level down.