Core Matter

Topic · Updated 12 August 2026

What does a humanoid robot cost to build?

The bill of materials, subsystem by subsystem: what each layer costs as a share of the whole, which layers have a domestic alternative, and what the public filings say about price.

40–60%

Actuation share of humanoid BOM cost

Third-party

65%

Domestic content the FCC rule requires, rising to 75% in 2029

Filing-grade

Rmb61bn

Unitree market capitalisation at IPO pricing, 6 August 2026

Filing-grade

$25,500

EngineAI PM01 entry price, mid-tier humanoid

Company-claimed

The current answer

  • 01Actuation is 40 to 60 percent of humanoid bill-of-materials cost, which makes it the layer that decides almost every other economic question about the robot. Core Matter estimate, August 2026, built from factory quotes and vendor pricing.
  • 02No other single subsystem comes close to actuation's share, so a cost curve for humanoids is mostly a cost curve for actuators. Compute, perception, battery cells and structure divide the remainder.
  • 03The FCC's July 2026 Covered List rule applies the Buy American test at 48 CFR 25.101(a): manufactured in the United States with domestic components above 65 percent of total component cost, rising to 75 percent in 2029. It is a bill-of-materials threshold rather than a country-of-origin ban.
  • 04The actuation share makes that threshold unreachable while actuators are foreign. At 40 percent of component cost and fully foreign, the highest domestic share a finished robot can reach is 60 percent. At 60 percent, the ceiling is 40 percent. Both fail 65 percent before a single other part is counted.
  • 05Unitree priced its STAR Board IPO at Rmb150.80 per share on 6 August 2026, implying roughly Rmb61bn of market capitalisation, about US$8.6bn. That is the first public price carried by a profitable humanoid maker, and the first real pricing event for embodied AI as an asset class.
  • 06Published humanoid prices span roughly two orders of magnitude by tier. EngineAI's PM01 starts at US$25,500, priced alongside AgiBot's A2 and Dobot's Atom, against Unitree's sub-Rmb100,000 developer tier at the low end and Droidup's Moya biomimetic humanoid at US$173,000 at the high end. All company-claimed.

The domestic-content ceiling, by actuation share

The arithmetic of the FCC test when the actuators are foreign-made. Neither end of the actuation range clears the threshold.

Actuation share of BOMMaximum domestic share reachableClears 65%?
40% (low end of range)60%No
50% (midpoint)50%No
60% (high end of range)40%No

The range does not need resolving to answer the question. No actuation cost share within it clears the test while the actuators are foreign-made. The rule splits domestic from everything else, with no allied tier, so Japanese and Taiwanese sourcing fails identically to Chinese sourcing.

Published humanoid prices by tier

Entry prices as published or reported, August 2026. All company-claimed unless stated.

Maker and modelPriceTierProvenance
Unitree, developer tierUnder Rmb100,000DeveloperCompany-claimed
EngineAI PM01US$25,500MidCompany-claimed
AgiBot A2Comparable to PM01MidCompany-claimed
Dobot AtomComparable to PM01MidCompany-claimed
UBTech UWorld companionAbout US$30,000Consumer companionCompany-claimed
Droidup MoyaUS$173,000BiomimeticCompany-claimed

Price tier and capability are weakly correlated. The developer tier exists to seed a research market rather than to do work, which is visible in the revenue split under Deployment reality.

The research behind this answer

Every Core Matter piece that contributed to the figures above, newest first, with what each one added.

  1. Report

    12 August 2026

    Unitree's IPO roadshow, translated: what Wang Xingxing told 360 investors

    Put a public price on a profitable humanoid maker, and disclosed the actuator architecture and revenue split behind it.

    Read it → · Data page →

  2. Newsletter

    4 August 2026

    Home robot economics 101: what 1X, Weave and Tau cost to run

    Took the BOM question into the home tier, where the buyer pool and the subsidy structure change the answer.

    Read it →

  3. Newsletter

    30 July 2026

    The FCC did not ban Chinese robots. It set a bill-of-materials test

    Walked the humanoid BOM layer by layer against the 65 percent domestic-content test, and established the actuation ceiling arithmetic.

    Read it →

  4. Report

    17 June 2026

    17,000 humanoids shipped last year. Almost none went to work.

    Established the shipment base and showed that vertical integration into motors, reducers and encoders buys iteration speed rather than only unit cost.

    Read it → · Data page →

  5. Newsletter

    6 June 2026

    NVIDIA's open humanoid runs on Chinese components

    A component-level read of a US reference design, showing where the parts actually come from.

    Read it →

  6. Newsletter

    28 May 2026

    Humanoid's two-German-supplier bet against the Tesla playbook

    The asset-light case: what changes in the BOM when a maker buys actuators and assembly rather than building them.

    Read it →

What this answer cannot cover yet

  • 01The subsystem-level cost model covering all eleven subsystems is built from factory quotes and is not yet published in full. The published figures above are the headline shares.
  • 02Actuation's 40 to 60 percent range is wide, and other analysts put it higher. Resolving that spread is the subject of the actuator primer in progress.

Related topics

How to read the figures

Every figure on this page is labelled by provenance. Filing-grade means it comes from a regulatory filing. Company-claimed means the company stated it. Third-party means a tracker, retail listing, teardown, or Core Matter estimate. This page is updated as new research lands rather than republished, and the date at the top is the last time a figure on it moved.

Research access

This answer moves when new work lands

The research that updates this page is published as it is finished, built from company filings, datasheets and direct factory pricing.

← All topics