Meta's Humanoid Runs on a Chinese Bill of Materials — and the US Robot Firewall Only Reaches the Robot

The July 28, 2026 Covered List addition reads as a nationality test and works as a geography test. The FCC's entry covers \"foreign-produced advanced robotic devices,\" and the public notice states the newly covered equipment is identified \"by place of production, not by entity.\" A robot assembled in Shanghai is covered whether it belongs to Meta, Fourier or a Chinese state-linked manufacturer; a robot assembled in Ohio is not covered even if every joint inside it shipped from Suzhou. The firewall taxes the assembly step, not the bill of materials.

The gap between instruments is where the money moves. The FCC's fact sheet states the action \"does not affect sales to, or use by, the federal government,\" so Washington's own determination leaves federal purchasing wide open. The American Security Robotics Act, announced by Senators Tom Cotton and Chuck Schumer on March 26, 2026 with a House companion from Representative Elise Stefanik, would close that door by barring federal agencies from buying or using Chinese humanoids and unmanned ground vehicles, with a narrow carve-out for military and law-enforcement research that cannot exchange data with China. Beyond introduction, the bill's status is not established by the sources reviewed here, so treat it as pipeline rather than law.

Actuators are where the economics live. Schaeffler puts actuators at roughly half the value of a humanoid's bill of materials, and Hyundai Mobis puts them near 60% of material cost. Chinese harmonic reducers sell at 40% to 60% of Japanese equivalent prices, and analysts cited in Chinese media put the resulting gap starkly: building an Optimus without Chinese suppliers would cost about ¥131,000 against ¥46,000 with them, against Tesla's ¥20,000 target. Clean sourcing roughly triples the bill

The margin evidence is where the price war shows up. Leader Harmonious Drive Systems lifted 2025 revenue 47% to about ¥569 million and more than doubled net profit to roughly ¥124 million, a net margin near 22%, on 425,158 reducers sold, with capacity targeted at 1.44 million units a year by end-2026 and a 30% to 40% share of China's harmonic reducer market. Harmonic Drive Systems went the other way: operating profit of about ¥7 million on ¥55.6 billion of sales in the year to March 2025, recovering to a 4.3% operating margin in the year to March 2026, with guidance of ¥68.0 billion of sales and ¥6.2 billion of operating profit for the year to March 2027. Chinese reducer economics run at roughly eight times the incumbent's net margin

That gap is what makes the humanoid option expensive on the wrong side of the ledger. Leaderdrive trades on roughly 377 times trailing earnings on the twelve months to June 2026, and Unitree on about 430 times trailing earnings on trailing revenue of ¥2.35 billion, while Harmonic Drive Systems changes hands at about 7.8 times enterprise value to sales. The market pays a premium for cost leadership it already expects to erode

Meta's own accounts show how far the humanoid programme is from being a business. Third-quarter 2026 revenue was $64.60 billion with $12.99 billion of operating income, while Reality Labs — the unit housing the robotics effort — turned $443 million of revenue into a $4.13 billion operating loss, with full-year 2026 Reality Labs losses guided to $18–19 billion. Capital expenditure of $33.94 billion in the quarter now exceeds the $32.79 billion of operating cash flow, and 2026 spending is guided to $135–145 billion. No robot line appears anywhere in it

What would turn the prototype into something financial: a formal agreement with Fourier, an FCC or Department of War conditional approval for a US-bound product, and a second prototype that reports measurable task performance. Until then this is an engineering milestone with a strategic cost attached — Meta's software stack and Nvidia's compute travel to Shanghai, while the finished robot cannot travel back without clearance. Prototype speed is the product; the commercial path is the risk

Read the chain end to end and the economics separate cleanly by layer. Compute and the model layer earn extraordinary margins and carry extraordinary multiples. The mechanical layer is where the Chinese cost curve bites: a Chinese reducer maker earned a net margin near 22% in 2025 while the Japanese incumbent earned a 2.7% net margin on far larger sales. The platform layer is loss-making and capital-hungry, with UBTECH still showing a trailing net loss of ¥597 million and Unitree's IPO leaving it on roughly 430 times trailing earnings after a 460% debut. Hardware margins are real; hardware multiples are borrowed from the option value of the robot

Over the next two quarters, the evidence arrives in a specific order: hardware milestones first, regulatory decisions second, order books third.

Over one to three years, three structural forces decide the outcome: whether the actuator cost curve can be rebuilt outside China, whether US onshoring actually lifts Chinese component volumes, and whether the compute layer keeps selling to both coalitions at once.

Fact, inference and speculation should be kept apart. Fact: Meta built a prototype humanoid with a Shanghai supplier in about three months, and on July 28, 2026 the FCC closed the US authorization door on foreign-produced humanoids while leaving components untouched. Inference: the rule relocates assembly rather than supply, so Chinese component makers gain demand while Chinese finished-robot brands lose US market access — which places the actuator trade upstream of the robot trade. Speculation: the gap between Meta's software and its Chinese hardware is a negotiating position rather than a hard dependency, and it breaks the moment either side signs something American. Own the joints and the compute, not the robot