Humanoid Robot Mass Production Race Heats Up as NdFeB Magnets Emerge as Next Supply Chain Bottleneck

Republished by AIC Engineering. All rights belong to the original publisher; see Source below.

The year 2026 is being hailed as the first year of mass production for humanoid robots, with Tesla announcing its Optimus production line will replace Model S/X lines with an annual capacity of one million units, triggering a wave of automotive suppliers including Aptiv and Valeo entering the embodied intelligence sector. However, before large-scale orders materialize, a supply bottleneck for neodymium-iron-boron (NdFeB) permanent magnets—a critical upstream material—is raising alarm. A single humanoid robot consumes roughly 4 kilograms of high-performance NdFeB, double that of a new energy vehicle. With rare earth separation and refining highly concentrated in China and capacity expansion constrained by capital costs, yield ramp-up challenges, and scarce talent, NdFeB could become the chokepoint restraining production if robot shipments reach the million-unit level in the coming years. This dynamic suggests profit allocation across the supply chain may tilt toward upstream magnet manufacturers.

Key Elements

When the roar of Tesla's Optimus production line begins to drown out the Model S and Model X lines, a supply chain restructuring centered on "embodied intelligence" will be quietly unfolding across the automotive industry.

2026 is being viewed as the inaugural year for industrial deployment of embodied intelligence, with humanoid robots accelerating their entry into automotive production lines. Tesla announced during last week's first-quarter 2026 earnings call that the company will "transform toward a future centered on AI, robotaxis, and humanoid robots." Tesla's first-generation robot production line will deliver capacity for one million units and replace the existing Model S and Model X production lines. Yet behind the rapid iteration of software algorithms, a physical hard constraint is surfacing—the key to how many humanoid robots can be built may lie hidden in the small magnets inside their joints.

Automotive Suppliers Pivot En Masse to Embodied Intelligence

At this year's Beijing International Automotive Exhibition, the scale of embodied intelligence technology displays was unprecedented. A Yicai reporter observed on site that Aptiv, Versigent, Valeo, Horizon Robotics, and several other automotive suppliers are now betting heavily on the embodied intelligence track.

Yang Xiaoming, President of Aptiv China and Asia Pacific, told Yicai: "Embodied intelligence perception and automotive intelligent driving perception are the same. Automotive companies' solutions can be applied in the humanoid robotics field. We are collaborating with embodied intelligence companies, hoping to leverage our massive system within the automotive ecosystem to provide technical solutions for deploying humanoid robots in industrial and other settings."

In visual perception, Aptiv's radar-vision integrated perception system PULSE can be used not only for close-range environmental sensing in vehicle platforms but can also extend the same fusion output capabilities to the robotics domain. Yang noted that the robot market's scale and future growth potential are enormous, emphasizing that humanoid robots should not remain confined to performing kung fu or appearing in films—real-world deployment in manufacturing scenarios holds far greater meaning. He stressed that while leveraging traditional automotive suppliers' strengths is important, technical adjustments are also necessary because humanoid robot manipulators demand higher perception capabilities and require force feedback, making it far from a simple copy-paste solution.

Qi Song, Global Vice President and China President of Versigent, also noted that humanoid robots and automobiles share many key technologies—both require an efficient nervous system, for example. Migrating experience accumulated in the automotive industry to emerging markets could create new growth engines. Valeo announced the establishment of a "humanoid robot training center" in Nanjing and publicly showcased for the first time at the auto show a portfolio of automotive-grade hardware and software products adaptable for humanoid robots and automated guided vehicles (AGVs). Qianxun SI also extended its spatiotemporal intelligence solutions from the automotive sector into embodied intelligence, launching its first "Embodied Spatiotemporal Brain."

Despite the cross-sector enthusiasm, large-scale orders have yet to materialize. Du Qian, Head of China Industrial Technology Research at Goldman Sachs, told Yicai in an exclusive interview that "most supply chain companies hold an optimistic, forward-looking view of the humanoid robot industry's prospects," but the supply chain has yet to see substantive public orders. She pointed out: "Taking industrial-grade humanoid robots as an example, we believe volumes need to reach the tens of thousands of units or above to form a supply chain system of meaningful scale."

The Underestimated "Magnet King" Bottleneck

While industry attention has been almost entirely consumed by VLA (Vision-Language-Action) models, world models, and end-to-end architectures, a physical hard constraint is being severely underestimated.

According to a DeepTech report, the factors determining humanoid robot production ceilings extend beyond algorithms and computing power to include a small magnet tucked inside each joint—neodymium-iron-boron (NdFeB) permanent magnets. Known as the "Magnet King," this rare earth permanent magnet material boasts the highest magnetic energy product of any commercialized magnetic material currently available.

Taking the Tesla Optimus Gen2 configuration as an example: 16 rotary actuators, 14 linear actuators, and two dexterous hands each equipped with 6 coreless motors total roughly 42 motors. At an estimated 3.5 to 4 kilograms of high-performance NdFeB per robot, a single robot consumes twice the high-performance magnetic material of a new energy vehicle.

Humanoid robots impose three hard requirements on magnets: high energy density, high coercivity (resistance to demagnetization at high temperatures), and mass-production consistency. Currently, the only commercialized material capable of satisfying all three simultaneously is NdFeB. Ferrite magnets offer energy density an order of magnitude lower; samarium-cobalt approaches comparable energy density but its price and brittleness cannot support mass production; non-rare-earth motors require four times the volume at equivalent power and simply cannot fit into a humanoid robot's knee joint.

More critically, magnet performance obeys crystal structures and physical laws, not Moore's Law. Over the past 40 years, the maximum magnetic energy product of NdFeB has improved from 30 MGOe to roughly 55 MGOe, with gains averaging less than 30% per decade. Engineers face severe physical constraints: reducing magnet material by 10% means insufficient motor torque density, triggering a chain-reaction amplification of overall machine weight; too much magnet material leads to core saturation and faster battery drain.

The "Inverted Pyramid" Risk in the Supply Landscape

Examining the mass production timeline, Nomura data indicates that Tesla Optimus Gen3 completed design freeze in the first quarter, with a full-year target of 60,000 to 80,000 units; China's six leading companies combined could reach shipments in the 110,000 to 200,000 unit range for 2026.

Adamas Intelligence stress tests show that if humanoid robots achieve large-scale deployment over the next 15 to 20 years, the required expansion multiples for NdFeB would reach dozens or even hundreds of times, far exceeding the capacity expansion needs for metals like copper, nickel, and lithium.

Yet the supply elasticity of the rare earth industry is extremely limited. The global rare earth supply chain exhibits a pattern of "fragmented at both ends, concentrated in the middle": mining and magnet manufacturing are relatively replicable, but China accounts for roughly 90% of the intermediate separation and refining segment. A separation and refining line with annual capacity of several thousand tons of oxides typically requires capital expenditure of USD 500 million to USD 1 billion, along with multi-year yield ramp-up periods and highly scarce process engineers. Over the past two decades, Western rare earth industries have attempted reconstruction multiple times but have often faltered midway, constrained by capital costs, yield challenges, and environmental investment requirements.

2026 to 2027 represents a relatively moderate transition period, with global humanoid robot demand for NdFeB accounting for only a single-digit percentage of the market. But by 2028 to 2030, if leading manufacturers realize even half of their capacity plans, demand will surge to tens of thousands of tons, and pricing power will begin shifting from downstream to upstream. Post-2030, if annual shipments reach the million-unit or even ten-million-unit level, NdFeB supply will become a direct constraint on robot mass production.

This dynamic is reshaping the strategic positioning of companies along the supply chain. China-based high-performance NdFeB manufacturer JL Mag Rare-Earth established a humanoid robot magnetic components business unit in early 2025, led directly by the CEO; Ningbo Yunsheng's supply to Agibot has entered mass production; Zhong Ke San Huan is cited in multiple brokerage research reports as one of the joint motor suppliers for Tesla's Optimus Gen2. Upstream magnet manufacturers are beginning to extend downstream into magnetic assemblies, rotors, and even complete motor modules, displaying a classic vertical integration move.

As Siemens China Chairman Xiao Song remarked, one challenge facing the industrial deployment of embodied intelligence is that data across the entire industry remains scarce. But beyond data, the physical limits of the hardware supply chain are equally defining the boundaries of this emerging industry. As software delivers diminishing marginal improvements and hardware's marginal progress slows further, the factor determining the industry's ceiling will gradually shift from the software side to the hardware side. Looking back a few years from now, the answers to how many humanoid robots are ultimately built and at what price they sell may lie not just in software and chips, but also in those dozens of grams of black magnets inside each joint.

Source

• finance.biggo.com (2026-04-27)
• Original article: Humanoid Robot Mass Production Race Heats Up as NdFeB Magnets Emerge as Next Supply Chain Bottleneck
• https://finance.biggo.com/news/ySk7z50B2jrwCtglFXMG