Electric won the humanoid joint, and the magnet became the risk

The day hydraulics left the humanoid mainstream

Fact On Apr 16, 2024, Boston Dynamics retired its hydraulic Atlas, a robot that had started in 2013 as a DARPA-funded project. A day later it showed an all-electric Atlas aimed at commercial work, with trials planned at parent Hyundai's facilities.1

Fact By January 2026 the electric Atlas had a dedicated actuator supplier. Hyundai Mobis said at CES on Jan 7, 2026 that it would supply actuators for the next-generation Atlas in line with its mass-production schedule, and that actuators make up more than 60% of a humanoid's material cost.2

Fact The parent group has scaled the plan since. Hyundai Motor Group has said it will deploy tens of thousands of robots over the next few years as part of a $26 billion U.S. investment, and it announced a North American robot factory with annual capacity of 30,000 units.2

Our read When the best-known hydraulic humanoid switched, the architecture debate for full-size bodies was largely over. For anyone selling robots, that changes the pitch. A buyer comparing three electric humanoids cannot see a difference in actuation type, so it stops being a selling point and becomes a question about sourcing, service and cost.

Hydraulics keep a foothold in the hand

Fact Sanctuary AI is the main exception. In December 2024 it showed 21-degree-of-freedom hands driven by miniaturized hydraulic valves, and said the approach gives an order of magnitude higher power density than cable and electromechanical systems. The company said it had tested its valve actuators over 2 billion cycles without leakage or degradation.89

Our read The hand is where force has to fit into the smallest volume, so it is the one place where a non-electric design still has a technical argument. These are company claims about a prototype. A buyer evaluating a hydraulic hand should ask for field service data before treating the cycle figure as a maintenance assumption.

The motor is cheap to make and expensive to source

Fact On Tesla's earnings call on Apr 22, 2025, Elon Musk said Optimus production had been affected by China's export controls on rare-earth magnets used in its actuators. Tesla was seeking an export licence, and Musk said China wanted assurances the magnets would not be used for military purposes.3

Fact China's April 4, 2025 controls require licences for seven medium and heavy rare earths, including terbium and dysprosium, and cover NdFeB magnets that contain them. On Oct 9, 2025 Beijing added five more elements and extraterritorial rules, then suspended that second round for one year after the U.S.-China trade truce. The April controls stayed in force.4

Fact As of early November 2026, the suspension is due to expire on Nov 10, 2026. Mining Technology reported in April 2026 that the options in front of Beijing were an extension, a selective reinstatement or a full reimposition. Oxford Economics wrote in August 2026 that China accounted for 60% of magnet rare-earth mining, 91% of refining and 94% of permanent-magnet production in 2024.56

Fact Supply outside China is growing but from a small base. Mining Technology, citing Bloomberg Intelligence, reported that non-Chinese NdPr output is expected to grow 4.4 times between 2024 and 2030, yet the world still faces a 36% NdPr shortfall by 2030. It cited GlobalData figures putting China at 69.2% of global rare-earth mine output in 2025.5

Our read A brushless motor is a mature, cheap part. The magnet inside it is not a free-market commodity. Any humanoid sold into a multi-year contract carries the licensing risk of its magnets, and the Nov 10 deadline means enterprise buyers will ask about it this quarter.

Washington is building a second source

Fact In July 2025 the U.S. Department of Defense agreed to buy $400 million of preferred stock in MP Materials and to lend it $150 million for heavy rare-earth separation, according to the company's 8-K filing. The Pentagon also guaranteed a floor of $110 per kilogram for NdPr for 10 years and agreed to buy, or allow resale of, the output of a planned second U.S. magnet plant for 10 years after it starts operating.7

Fact The terms run both ways. The filing sets a minimum annual EBITDA of $140 million for the new magnet facility once it reaches full production, and gives the Pentagon 30% of any NdPr sales price above $110 per kilogram once that plant is at full capacity.7

Our read That plant does not exist yet at scale, so for now a non-Chinese magnet is a promise more than a part number. Still, defence-adjacent and public-sector buyers will start asking for a domestic magnet roadmap. A humanoid vendor that can name a non-Chinese magnet source, even a future one, answers a question its competitors cannot.

Our read The price floor also tells a buyer something about cost. A guaranteed $110 per kilogram floor for a U.S. source suggests a domestic magnet will not be the cheap option. Vendors who promise a non-Chinese supply chain should expect a higher actuator cost and price the contract accordingly.

Frames and the magnesium question

Fact Lightweight metals carry a similar concentration. The U.S. Geological Survey estimates China produced about 950,000 of the world's roughly 1.1 million tonnes of primary magnesium in 2025, and notes that the only U.S. primary smelter, in Utah, stopped production in 2022. Its operator filed for Chapter 11 in September 2025.11

Fact Washington is funding alternatives here too. USGS reports that the U.S. Department of War awarded $19.6 million under the Defense Production Act in February 2024 to a California company producing magnesium from brines, which met its purity target at pilot scale in July 2025 and announced plans in December for a commercial plant in Arkansas.11

Our read Magnesium matters less than magnets today because most humanoid frames still use aluminum. But every vendor chasing weight for longer battery life will look at it, and when they do they inherit the same China dependency. Supply teams should track it before a design team commits to it.

Vertical integration as a margin strategy

Fact Unitree's prospectus lists motors and gear reducers among its in-house core components. In the first three quarters of 2025 its gross margin was 59.5%, with humanoids at 62.9% and its component sales at 60.4%, according to a KrASIA report on the filing. Unitree reported 2025 revenue of 1.699 billion yuan and said it shipped more than 5,500 humanoids that year.1012

Fact The shift happened fast. Humanoids were 1.88% of Unitree's revenue in 2023, when quadrupeds made up 75.78%, according to the Global Times report on its IPO. Chinese investment banks estimated its post-listing valuation could exceed 40 billion yuan.12

Fact Western players are taking the opposite route and buying actuators as modules. Schaeffler's November 2025 deal with Neura covers planetary gear actuators rated up to 250 Nm for shoulders, elbows, knees and wrists.13

Fact Some U.S. players are integrating too. Figure says it brought key modules in-house for Figure 03, including actuators, batteries, sensors, structures and electronics, and that the new actuators run up to twice as fast with better torque density.14

Our read Both routes can work, but they produce different sales stories. Unitree can price low and still earn a 60% margin because it makes its own joints. A company buying modules from a Tier-1 has to win on software, service and the workflow it automates, because its hardware cost floor is someone else's price list.

Our read Unitree's numbers also show where China's magnet advantage turns into price. A maker that builds its own motors next to the world's main magnet supply carries less licensing risk on its home market than a Western rival importing the same magnets. U.S. buyers comparing quotes should ask what share of each vendor's price is exposed to export licences.