Hesai (HSAI), XPeng (XPEV) Build Humanoid Robot Joints In-House

XPeng (XPEV) and Hesai (HSAI) both told investors in late August 2026 that humanoid robot joint actuators are in short supply, and both said they are building their own rather than buying them. Hesai spoke as a seller and booked its first revenue from the parts; XPeng spoke as a buyer and said it now makes more than 85% of its robot's core components in-house.

The most expensive part of a humanoid robot is its joints

A humanoid robot's costliest component is not its chip. Every movable joint needs an actuation module that packages a motor, a reducer, an encoder and a controller together. Hesai's management said one humanoid may need more than 100 such modules, that actuation can account for more than half a robot's bill of materials, and that the margin profile on the business is currently around 40% [1]. Building these parts requires automotive-grade precision electromechanical manufacturing — motor windings, gear machining, batch consistency — the same production experience used to make LiDAR units or cars.

The established division of labor put design and assembly with the robot maker and joints with specialist reducer and actuator suppliers. What changed this quarter is who adds the next increment of joint capacity. Hesai turned it into a new segment sold to outside customers; XPeng locked it inside its own supply chain [1][2]. Neither company came from the robotics component industry, but both hold automotive-grade manufacturing capacity — which is why this may not stop with these two. Automakers and auto electronics suppliers in the same position face the same choice.

First RMB 45 million of revenue, and an 85% in-house ratio

Hesai's robotics and emerging business segment recorded its first revenue in the second quarter — RMB 45 million (about US$7 million), driven by early demand for actuation modules — against a segment operating loss of RMB 64 million. The company raised full-year 2026 revenue guidance for the segment from RMB 100 million to RMB 200-300 million [1]. Management described the part as scarce:

"Every robotics company is hunting for great joints, but they're hard to find. Joints are both critical and hard."

Six days later XPeng confirmed the same shortage from the buyer's side. Its risk disclosure states that supply of core components is currently scarce and may affect the early production ramp, while the company disclosed that more than 85% of IRON's core components are developed in-house, covering chips, controllers, motion modules and dexterous hands [2]. Management added that robots on the market are priced at roughly 2.5 to 3 times their bill of materials, against a 12.1% vehicle gross margin at XPeng in the same quarter [2].

The two moves run in opposite directions — one adds outside supply, the other withdraws its own purchasing — but point to the same result. The incremental output of the joint layer is being supplied by manufacturers who were not previously in that business.

Robot makers that buy their joints carry a worse cost structure

If modules account for more than half the bill of materials and suppliers price them at roughly 40% gross margin [1], a robot maker that does not build them gives that margin away. While finished robots still sell at 2.5 to 3 times their bill of materials [2], that gap largely determines who reaches profitability first, and it explains why XPeng chose to build rather than buy. For incumbent specialist module makers, the addressable buyer base narrows from all humanoid production to the share produced by companies that do not build in-house.

The evidence currently runs in one direction only. No merchant module supplier has disclosed any order loss, and Harmonic Drive Systems' fiscal 2026 revenue was about 17% below its fiscal 2023 peak [4], indicating humanoid demand has not yet reached the merchant tier's income statement. Most of Hesai's new actuation demand also comes from a customer connected to its own CEO, under a supply framework agreement whose annual cap was just raised from RMB 100 million to RMB 300 million [1], which limits how independent that demand is. The metric to watch: whether merchant module suppliers' revenue tracks total humanoid production or only the output of makers that do not build in-house.

Companies exposed to this change

• Leaderdrive (688017.SH): A specialist merchant supplier of harmonic reducers and joint modules whose revenue depends entirely on outside procurement by robot makers; its addressable volume narrows as in-house ratios rise. It trades at about 88 times sales [3], and has disclosed nothing indicating its orders have been affected.
• Harmonic Drive Systems (6324.T): A global precision reducer maker exposed to the same rule change as Leaderdrive. Its fiscal 2026 revenue was about 17% below the fiscal 2023 peak [4], and humanoid demand is not yet visible in its results.
• UBTECH Robotics (09880.HK): A humanoid robot maker that does not build its own joints and therefore pays market prices for the largest single item in its bill of materials. Its fiscal 2025 group gross margin was 36.3% [5], below the roughly 40% margin Hesai reports on its actuation business.

Sources

[1] Drillr - Hesai Group (HSAI) - 2026-08-18 - FY2026 Q2 earnings call

[2] Drillr - XPeng Inc. (XPEV) - 2026-08-24 - FY2026 Q2 earnings call

[3] Drillr - Leaderdrive (688017.SH), UBTECH Robotics (09880.HK) - 2026-08-25 - company snapshot valuation data

[4] Drillr - Harmonic Drive Systems (6324.T) - 2026-08-25 - financial statements

[5] Drillr - UBTECH Robotics (09880.HK) - 2026-08-25 - financial statements

This is only meant to surface industry changes and companies you may have overlooked - it is not a stock recommendation.