MinebeaMitsumi: The Position Japanese Precision Component Manufacturers Are Aiming for in the Era of Physical AI
The bizarre scene unfolding at the Las Vegas venue
January 2026, at the CES venue in Las Vegas. At one Japanese company's booth, a five-fingered robotic hand was lifting a 5kg dumbbell with just its little finger (according to an on-site report by Nikkei CrossTech). In another demonstration, it used cameras to read the hand movements of spectators standing in front of it and reproduced them almost in real-time. In yet another demo, it was using strings attached to each finger to move a marionette at high speed.
What is inside this hand? Each hand contains 107 bearings, micro-actuators with reduction gears, and strain sensors. The company bundling all of these together is MinebeaMitsumi (6479), a "Sogo (integrated) precision component manufacturer" headquartered in Miyota, Nagano Prefecture.
This is their first time exhibiting at CES. Why is a miniature bearing manufacturer with 1.6 trillion yen in sales and over 60% of the global market share now pushing robotic hands to the forefront? Following the reasons reveals the "true bottleneck of the Physical AI era" and the position that Japanese precision component manufacturers are aiming to secure.
To begin with, what kind of business is MinebeaMitsumi?
Extracting the business segments from their official IR, it looks like this (FY2026/3 results, announced May 12).
Precision Technologies (PT) — Ball bearings, HDD pivots, aircraft screws. Sales of 281.1 billion yen, operating profit of 62.2 billion yen (operating profit margin of 22.1%).
Motors, Lighting & Sensing (MLS) — Fan motors, HDD spindles, stepping motors, sensors. Sales of 456.5 billion yen, operating profit of 26.9 billion yen (5.9%).
Semiconductors & Electronics (SE) — Semiconductors, optical devices, mechanical components, power supplies. Sales of 590.2 billion yen, operating profit of 26.6 billion yen (4.5%).
Access Solutions (AS) — Automotive (key sets, door latches, door handles, etc.), industrial equipment components. Sales of 332.2 billion yen, operating profit of 17.0 billion yen (5.1%).
Combined, this totals 1.6643 trillion yen in sales and 103.9 billion yen in operating profit. This represents a year-on-year increase in both revenue and profit of +9.3% and +10.1%, respectively.
What is noteworthy is that the PT segment is generating an operating profit margin of 22%. What does this mean? Their core business of miniature and small-diameter ball bearings holds approximately 60% of the global market share, and their HDD pivot assemblies hold approximately 80% (both according to company research). The position of being an "exclusive supplier of irreplaceable ultra-precision components" is what creates this structural high profitability.
And onto this, the companies added through recent M&A are layered: Mitsumi Electric in 2017 (sensors, semiconductors, connectors), U-Shin in 2019 (automotive door locks, handles), ABLIC in 2020 (analog semiconductors), Minebea Power Device in May 2024 (acquisition of the former Hitachi Power Device was decided in November 2023, completed and renamed in May 2024), and Minebea Linear Motion in October 2025.
In other words, the current MinebeaMitsumi is not just a "bearing company." It is a company that bundles and sells eight categories (what the company calls its "Eight Spears"): bearings, motors, sensors, semiconductors, power supplies, connectors, high-frequency components, and software. The reason the company officially writes "Sogo" (integrated) using characters meaning "mutual combination" rather than the standard term for "comprehensive" is that it is a term intentionally coined to mean not just holding multiple businesses, but "combining them to create something new."
This is the foundation of the company. From here on, it connects to the story of Physical AI.
Why are "precision components" the key battleground for Physical AI?
Since around 2024, the focus of the discussion on generative AI has been shifting little by little. LLMs like ChatGPT have become capable of "thinking." Visual recognition and multimodal understanding have also reached a practical level to some extent. The next step is for AI to enter the stage of moving in the real world — this is the context of "Physical AI," which NVIDIA CEO Jensen Huang has been repeating.
Large-scale investments are flowing into humanoid robots all at once. Figure AI, Tesla Optimus, China's Unitree, SoftBank's agreement to acquire ABB's robotics business — the flow over the past year has had an extraordinary sense of speed.
However, if you look closely, the main battlefield is changing. The bottleneck on the software side has become smaller. The remaining difficult challenge is the "body" side.
For a humanoid to truly become a "usable" entity in the world, it must satisfy all of the following as a machine: joints that can exert power and precision equal to or greater than a human at the same size, durability that allows it to operate for hours without breaking, power efficiency that makes battery operation viable, and a cost that allows for mass production (this is the most difficult). Furthermore, from this point on, the problem cannot be solved by "rewriting software." This is a domain that can only be tackled by companies with a long history of precision machining and mass production.
In other words, what truly makes the difference in physical AI is not the AI model itself, but whether one can achieve "precision actuation of fingers and joints at mass-production costs." And there are several Japanese companies that can compete globally in this area. Among them, MinebeaMitsumi is the player that is most clearly "coming to compete with a full suite of technologies."
What is inside the robot hand at CES 2026?
I will organize the component parts by cross-referencing the official press release and on-site reports (EE Times, Nikkei CrossTech).
Micro-actuator with reduction gear: The size is only 13mm x 19.4mm x 60.4mm. Developed jointly with Harmonic Drive Systems (7276). It features two types of reduction ratios, 1/100 and 1/30, simultaneously, achieving both "power (high torque)" and "speed."
107 bearings: The key to supporting each finger joint. The company's know-how in miniature bearings is applied directly here (the company's CES exhibition and financial results materials state 107. Note that some parts of the product page mention "85 small-diameter bearings," so there may be differences in definition or specifications).
Strain sensors (built into fingers): Recognize whether the object being touched is hard or soft, and change the grip through feedback control.
Small 6-axis force sensor MMS101: Only 9.6mm in diameter. It drastically miniaturizes conventional 6-axis force sensors (typically around φ80mm, with the smallest being φ17mm) and achieves a price point half that of conventional products.
Other components, such as motors, connectors, and waterproof parts, are all from the company's existing lineup.
What is noteworthy is that almost all of these can be sourced within the company's own group. Normally, when trying to build a robot hand, one has to order from different component manufacturers, ensure compatibility, and assemble them. MinebeaMitsumi can create a situation where "when you order from a bearing manufacturer, the motors, semiconductors, and sensors all come along with it." This is the meaning of "Sogo" (integration).
And MinebeaMitsumi's aim appears to be not the finished robot itself, but to take a position as a supplier of parts and modules for humanoids. Various reports have mentioned commercialization as a "general-purpose finger module"—in other words, they are trying to take a position where they can tell manufacturers around the world who are building humanoids, "You can just use our modules for the fingers."
This is the same structure as "Qualcomm chips" and "Sony image sensors" in the smartphone industry. Instead of competing with finished products, they become the standard component supplier for the entire industry. Once established, the lock-in effect is powerful.
Five pillars: It's not just about humanoids
MinebeaMitsumi has clearly stated its "five pillars" as future growth drivers (from the CES 2026 exhibition release on December 16, 2025).
One is AI server products (fan motors, HDD spindle motors, pressure sensors, optical connectors). Second, humanoid robot products (bearings, ball screws, semiconductors, frameless motors). Third, parts for LiDAR for fully autonomous driving (cartridge bearings, BLDC motors, coils, power inductors). Fourth, products for drones (bearings, brushless motors, GNSS antennas, optical markers). Finally, integrated products such as automotive wing handles and aperture actuators for small camera modules.
What I personally think is important here is the first one, AI server products. While humanoids have great "dreams," it takes time until mass production. On the other hand, fan motors for AI servers are selling at this very moment.
The official financial summary states: "Sales of our mainstay ball bearings increased due to steady demand for servers for data centers and demand for aircraft" (PT segment explanation). "Sales increased mainly due to increased demand for fan motors" (MLS segment explanation).
In other words, behind the future showcase of the robot hand, they are capturing the computational demand of generative AI itself through data center cooling. For the few years until humanoids become mainstream, AI server products will generate cash, and during that time, they will nurture products for humanoids, autonomous driving, and drones. As a business structure, it is quite well-constructed.
Issues that still remain: What to look for from here on
Reading this far, only "good stories" are lined up, but whether it will truly be effective is another matter. As points of discussion, I will list the points I want to follow in the future.
First, the deep valley between PoC (prototyping) and mass production. There is a significant gap between showcasing a prototype of a robot hand at CES and mass-producing it in a state where it is 'adopted' by humanoid manufacturers around the world. What to watch for is whether announcements will emerge regarding 'specifically which manufacturers have adopted it, and in what quantities.' If it stops at the prototype stage, it will end up as nothing more than a PR event.
Second, price competition with Asian players. At CES 2026, many Asian robot hand-related companies, including those from China and South Korea, were exhibiting (DexRobot, Dyna Robotics, Galaxea Dynamics, LY iTECH, South Korea's Aidin Robotics, and others). While it is true that Japanese companies have a technological advantage in the field of precision components, price competition for general-purpose modules will become severe once other countries catch up in mass production costs. The question is whether the company's 'bundled sales through integration' can structurally protect its price advantage.
Third, FY2027/3: How to view the growth of the core business excluding special factors. Net profit for this fiscal year (FY2026/3) was a record high of 99 billion yen. However, this includes a special factor: the fair value valuation gain on financial assets (related to Toshiba shares) held by the company. The company's forecast for FY2027/3 net profit is 83 billion yen. On the surface, this is a profit decline, but it can also be presented as a forecast for profit growth on a core business basis, excluding special factors. Stock valuation from here will be a debate on how to weigh the 'disappearance of special factors' against 'core business growth related to physical AI'.
Fourth, the essential strength of the 'integration' strategy. Does the style of 'bundling multiple businesses to sell' truly become a competitive advantage? It is a common story in the history of manufacturing that rivals who specialize in individual components with individual optimization end up winning in both price and performance. On the other hand, in applications where 'integrated design of multiple components is effective,' such as robots and automotive parts, there is a high possibility that the integration strategy will work. Regarding this, it is necessary to watch how the preferences of users (the robot manufacturers) shift over the next few years.
An era where the 'invisible giant' is tested
MinebeaMitsumi is not a flashy AI company. It does not release products that reach general consumers directly like ChatGPT. The company's commercial tagline is that its products are 'secretly and thoroughly inside everything,' and from the perspective of an end user, it is almost non-existent.
However, if an era comes where humanoid robots walk the streets, work in factories, and provide services in homes, the things rotating their joints might be the miniature bearings founded in Itabashi, Tokyo in 1951 and refined in Karuizawa, Nagano since the 1960s, along with the 'integration' technology that bundles them together.
Behind the stories of flashy 'AI stocks,' keeping even one such 'player holding the physical layer' on the map in your head will raise the resolution of the theme of physical AI by one level. That is what I felt during this research. Before making an investment decision, it is, first of all, an interesting company as an industrial structure.
The figures and citations in this article are based on MinebeaMitsumi's official IR materials (FY2026/3 Financial Results Summary announced on May 12, 2026), the company's press releases (CES 2026 exhibition guide on December 16, 2025, and completion of the acquisition of Hitachi Power Semiconductor Device on May 2, 2024), as well as on-site reports from CES 2026 by EE Times Japan and Nikkei CrossTech. For global market share figures, please refer to the company's own research, such as 'Know MinebeaMitsumi in 3 Minutes'.
[Reference Links]
Primary Sources (Official IR/Press Releases):
FY2026/3 Financial Results Summary https://www.minebeamitsumi.com/corp/investors/disclosure/financial/p2026/summary04/
CES 2026 Exhibition Release https://www.minebeamitsumi.com/news/press/2025/1210636_20342.html
Humanoid Product Page https://product.minebeamitsumi.com/pickup/humanoidrobot/
On-site Reports:
EE Times Japan https://eetimes.itmedia.co.jp/ee/articles/2601/20/news074.html
Nikkei CrossTech https://xtech.nikkei.com/atcl/nxt/column/18/03455/011600020/
Video (Reference, filmed by third party, on-site at CES 2026):
Robotic Hands at CES 2026 (Mario Herger) https://www.youtube.com/watch?v=FI0tg7xh5Sk
(Vlox)
