A Logbook on Humanoid Robots/Physical AI: A Deep Dive into Structure, Companies, and Risks Based on the Current Situation in 2026
A Logbook on Humanoid Robots/Physical AI: A Deep Dive into Structure, Companies, and Risks Based on the Current Situation in 2026
This article delves into the structure, companies, and risks of humanoid robots/physical AI, based on the current situation in 2026.
Current Market Situation: 2026 is a Turning Point from "Proof-of-Concept to Mass Production"
Mass production of humanoid robots will begin in earnest.
Tesla Optimus: Initial production of Gen 3 will begin in summer 2026, with mass production starting in summer 2027. Approximately 10 million square feet (approximately 930,000 m²) of factory space will be secured specifically for robot manufacturing. Target price is under $20,000, with external sales to other companies beginning in 2026.
Figure AI: Announced the home-use Figure 03 in October 2025. Selected as one of TIME magazine's "Best Inventions of 2025." Partnered with BMW.
Price disruption by Chinese companies: Unitree's G1 ($13,800) and R1 ($5,900) are priced exceptionally low. UBTECH, AgiBot, and NOETIX Robotics report achieving annual shipments of 10,000 units.
Market forecasts: Goldman Sachs has revised its long-term forecast upward by six times, and Morgan Stanley presents an extremely bullish scenario of 1 billion units in operation by 2050. Fuji Keizai predicts a global market of 3.5 trillion yen by 2035.
However, a calm assessment reveals two realities:
While the videos are impressive, actual factory operation is still limited. Even Tesla's Optimus was reported to have partially assisted with visitor handling at its Cybercab announcement event in October 2024.
Dissolution of the Dojo development team: In August 2025, Tesla disbanded the Dojo supercomputer development team, reallocating personnel as part of a strategic AI shift. Building the learning infrastructure is proving more difficult than expected.
Cost Structure of a Single Humanoid Robot and the Position of Japanese Companies
Tesla Optimus Gen 3 Specifications (Estimated)
Total 66 Degrees of Freedom in Both Hands (22 DoF x 2 fingers)
Equipped with Tesla's proprietary AI5 chip
Samsung OLED Face Display
Target Price: Under $20,000 (Mass Production)
Component Cost Structure (Industry Estimate) For an Optimus-class humanoid robot, actuators + gearboxes are estimated to account for 30-40% of the total cost. Each robot has 40-60 joints, and each joint requires a motor + gearbox + encoder + bearing + driver IC.
Therefore, with a target price of $30,000, the procurement cost of a single gearbox is limited to several thousand to 20,000 yen per joint. This is where Japanese companies need to compete.
Japan's Strengths and Weaknesses
Strengths: Gear reducers (Harmonic Drive, Nabtesco), precision bearings (MinebeaMitsumi, NTN, NSK), servo motors (Yaskawa, NIDEK), image sensors (Sony), torque sensors (Leptrino, Wakotec)
Weaknesses: AI-based models, VLA (Vision-Language-Action), integrated systems, software
Harmonic Drive Systems (6324): The absolute king of harmonic drive gears
Business Model
Main product: Harmonic Drive gear reducer: Proprietary technology that simultaneously achieves small size, light weight, and high reduction ratio
Global market share of approximately 50% (Competitors: Germany's Spinea, China's Green Harmonic Drive)
Used in industrial robots, semiconductor manufacturing equipment, surgical assistance robots, and humanoid robots
Ideal for small joints such as wrists, fingers, and elbows of humanoid robots (RV reducers are for large joints)
Strategic investment in humanoid robots The company plans to invest 27.5 billion yen in capital expenditures over the three-year period from 2024 to 2026, with approximately one-third (approximately 10 billion yen) allocated to humanoid robots. This includes establishing a mass production system at the Ariake Plant (Azumino City, Nagano Prefecture) and expanding the cross-roller bearing production at its subsidiary, Harmonic Precision (Matsumoto City). Aiming for sales of 10-20 billion yen in humanoid robot reducers by FY2026
Current Performance
FY2025 Results: Sales 55.6 billion yen, operating profit margin nearly 0%
Medium-Term Management Plan FY2026 Target: Consolidated sales 90 billion yen, operating profit margin 15-20%
FY2026 Q2: Consolidated operating profit significantly improved year-on-year (inventory reduction and improved marginal profit margin)
FY2026 Q3 Cumulative: Consolidated ordinary profit turned positive at 1.25 billion yen (compared to a loss of 70 million yen in the same period last year), full-year plan progress rate 83.8%
Structural Trends in 2025: The current medium-term management plan anticipates a recovery in orders due to increased capital investment driven by increased investment in automation, including robots, to compensate for the declining workforce, expansion of data centers, and increased demand for advanced semiconductors necessary for generational AI.
Points to Note
Stock prices are projected to rise significantly in 2025 due to the humanoid theme.
Chinese companies (such as Green Coexistence) are beginning to catch up.
At the current pace, there is a risk that the company will not achieve its medium-term sales target of 90 billion yen.
Nabtesco (6268): 60% global market share for RV reducers
Differentiation with Harmonic Drive Systems
Nabtesco's RV (Rotary Vector) reducers: Strengths include high torque and high rigidity, for large joints such as shoulders and waists of industrial robots.
Harmonic Drive Systems: Small and lightweight, for small joints such as wrists and fingers.
Harmonic Drive Systems has an advantage in Optimus-class humanoids (emphasis on miniaturization), while Nabtesco has an advantage in large humanoids for construction and nursing care.
Cumulative shipments and market position
Cumulative shipments of RV reducers exceed 7 million units
Approximately 60% global market share for industrial robots
JPX Nikkei 400 constituent stock
Positioning for humanoid robots It has a high probability of being adopted as a main component in large humanoid robots (such as Kawasaki Heavy Industries' Kaleido, which is over 180cm tall). However, in the trend towards miniaturization and weight reduction, Harmonic has the advantage.
THK (6481): A rare player that also develops the humanoid robot body itself.
Core Business
LM (Linear Motion) Guides: Over 50% global market share (basic components for industrial machinery)
Ball screws, cross roller bearings
Comprehensive manufacturer of precision mechanical components
Proprietary humanoid robot "SEED-Noid" and FRED
The humanoid robot "FRED" is equipped with high-performance actuators with high output performance and a flexible shock absorption mechanism, specializing in running performance. In a running test in May 2025, it recorded a world-leading speed of 13.8 km/h (3.8 m/s) for a humanoid robot.
The legs feature a uniquely designed link mechanism, achieving both a structure that absorbs shocks from the ground and high controllability. Real-time control controller operating with a 1ms cycle
Attracting attention with its exhibit at the Osaka-Kansai Expo
THK is unique among Japanese companies in that it "manufactures finished products in-house." It achieves comparable driving performance to Boston Dynamics using its own component technologies.
FANUC (6954): Core of the Physical AI theme
Rising to 4th place in Nikkei 225 contribution. While semiconductor and AI-related companies such as Advantest, Tokyo Electron, and SoftBank Group are among the top contributors to the Nikkei 225, FANUC ranked 4th. This reflects market expectations for physical AI (robotics x AI).
FANUC's Strengths
Approximately 20% global market share for industrial robots (World's top 4: FANUC, ABB, KUKA, Yaskawa)
Over 50% global market share for NC devices (machine tool controllers)
Expected recovery in orders from 2026 onwards following the "Mt. Fuji Robot Shock" (2023-24)
Connection to Humanoids: While FANUC itself doesn't manufacture humanoids, its design philosophy for highly productive automated cells (robots + machine tools) serves as a benchmark for ROI calculations when introducing humanoids. Existing industrial robots are so dominant that the question "Why humanoids?" remains unanswered within factories. This is both FANUC's strength and, conversely, a barrier for humanoids.
Sony Group (6758): Image Sensor Dominance
Approximately 50% global market share for CMOS image sensors
Dominant in smartphones, automobiles, and industrial applications
Depth cameras, ToF sensors, and high-sensitivity CMOS image sensors are essential for the "eyes" of humanoid robots.
Sony's CMOS sensors are used in all Tesla EVs, and the same structure makes their use highly likely in humanoid robots.
Aibo Asset: Sony has over 20 years of experience with the Aibo home robot, possessing know-how in small servos, sensor fusion, and learning-based behavior generation. However, scaling up to commercial humanoid robots has not yet begun.
Keyence (6861), Omron (6645): Vision Sensors
Keyence
World-leading FA sensors, with an astonishingly high operating profit margin of around 50%
3D vision, laser displacement sensors, code readers
Direct connection with domestic and international industrial robot manufacturers for robot vision
A highly successful company with a market capitalization of approximately 20 trillion yen
Omron
Development of collaborative robots "TM series"
Top domestic market share in control equipment for FA
Growth projected in the first quarter of the fiscal year ending March 2026, capitalizing on the recovery in semiconductor-related demand
Positioning of both companies: While they do not directly manufacture humanoid robots, they control the sensing and integration infrastructure for humanoid robots introduced into factories. Keyence, as a provider of factory-wide automation solutions, has the potential to become a major player even in the era of humanoid robots.
MinebeaMitsumi (6479): Bearings and Small Motors
Strength lies in the wide range of components.
60% global market share in ultra-small ball bearings.
Small motors (stepping, DC, BLDC)
Vertical integration from semiconductors, analog ICs, to optical components.
Supplies to aerospace, medical, industrial robots, and humanoid robots.
Potential for humanoid robots: 50-100 small bearings are required per robot. If Optimus expands to 100,000 units per year (Tesla's 2026 target), 1 million units (2028 target), and 10 million units (2030 target), a simple calculation shows a bearing demand of 500 million to 10 billion units. This is a potential market several times the total sales of Minebea.
Japanese Humanoid Robot Manufacturers
Kawasaki Heavy Industries (7012): Kaleido/Friends
Kaleido: Approximately 190cm tall, bipedal, powerful enough to lift heavy loads and robust enough to withstand falls. Friends: 168cm tall, slim body, designed to work alongside humans as a partner.
Intended for use in disaster relief and caregiving settings.
Kawasaki Heavy Industries is the world's 5th largest producer of industrial robots and a long-established company that began producing Japan's first industrial robot in 1969.
Toyota: T-HR4
Toyota announced its 4th generation humanoid robot, "T-HR4," in December 2025. It is 160cm tall, weighs 70kg, and boasts 1.5 times the work speed of a human thanks to improved torque servo modules. In field tests at its own factory, it achieved a 60% automation rate in assembly work. Leasing to external companies to begin in 2026, starting from 300,000 yen per month
Honda: ASIMO-X
A new project, "ASIMO-X," inheriting the technology of ASIMO, will launch in January 2026. It will be redefined as a mobile work robot, integrating with autonomous driving technology for electric vehicles. It features a dialogue system achieving 99.5% voice recognition accuracy.
Yaskawa Electric (6506)
One of the world's top four industrial robot manufacturers.
In 2025, it will acquire the humanoid startup "Donut Robotics" as a subsidiary.
While it doesn't produce finished products itself, it sets industry standards for servo motors and controllers.
Kawada Robotics, SoftBank Robotics, RT, Vstone:
Despite being a small-to-medium-sized enterprise, it has a proven track record in the research and education market.
11. AI Robot Association (AIRoA) and VLA-based Model
Members: The AI Robot Association (AIRoA) includes Toyota Motor Corporation, Hitachi, Ltd., and KDDI. In recent years, NEC and Fujitsu, which are major players in digital solutions, have also joined.
The Importance of VLA (Vision-Language-Action)
Traditional robot control: Position and speed are instructed by program.
VLA model: Visual information + natural language instructions → Direct generation of actions.
This eliminates the need for programming individual tasks, enabling general-purpose robots that can "move flexibly even in unknown environments."
Weaknesses of Japanese Companies
Compared to Figure AI's "Helix" (US), Google DeepMind's "RT-2," and China's Zhiyuan Robotics' "GO-1," Japanese companies lag behind in base models.
Due to inferiority in GPUs, datasets, and learning infrastructure, while hardware is strong, software cannot keep up.
Significance of AIRoA: Development of a "robot base model" that collects robot operation data and connects it to efficient learning has begun. They are attempting to catch up through data collection and standardization. However, they are far behind the capital power of the US and China.
Five Criteria for Mass Production and Implementation
Calm Judgment Criteria for Investing in and Engaging in Humanoid Robots:
Criteria | Points to Consider | Current Winners 1. Manufacturing Scale | Annual Shipments, Factory Area | Tesla, UBTECH, Unitree 2. Long Operating Time and Maintenance | MTBF, Parts Replacement System | Agility Robotics (Digit) 3. Field Operational Track Record | Major Customers, Implementation Examples | Figure (BMW), Agility (GXO, Amazon) 4. Safety and Regulatory Compliance | ISO 10218, Collaborative Robot Certification | Boston Dynamics, Japanese Industrial Robot Manufacturers 5. OS, VLA, and Development Platform | Learning Data, Software Ecosystem | Figure (Helix), Tesla, Chinese Companies
Realistic Implementation Scenario: The actual implementation of humanoid robots will begin not in homes, but in industrial applications where constraints are easier to define. It's no coincidence that many companies are first testing it in factories and logistics; it's because the work is easier to define, and safety and cost-effectiveness are relatively easier to assess.
Specific Investment Recovery Scenario
If Tesla Optimus achieves its target price, the investment can be recovered in 2-3 years compared to a worker with an annual labor cost of 4 million yen (assuming the target price is achieved).
Realistically, implementation will begin with 24-hour logistics warehouse picking, simple assembly processes, and cleaning.
Popular adoption in homes is predicted for the late 2030s to 2040s.
Risks and Points to Note (A Calm Perspective)
Overestimation of "Flashy Demos": Tesla Figure's demo videos are basically promotional material edited to show the best-case scenario. Whether it can be implemented in mass production, maintenance, safety, and profitability is another matter.
Price Offensive from Chinese Companies: The prices of the Unitree G1 ($13,800) and R1 ($5,900) directly impact the profitability of Japanese parts manufacturers. If reduction gears, motors, and bearings are manufactured domestically in China, it will put pressure on component sales for Japanese companies.
Risk of Harmonic Drive's Performance Failure: While the medium-term sales target was 90 billion yen (FY2026), actual sales for the fiscal year ending March 2025 were 55.6 billion yen, significantly below target. The timing of achieving the 10-20 billion yen sales target for humanoid robots depends on the mass production launch by customers; a delay in the start of mass production for Tesla Optimus in summer 2026 will also cause a delay.
Legal Regulations, Occupational Accidents, and Liability Issues: In the event of an accident involving a humanoid robot coming into contact with a person in a home or commercial facility, the scope of manufacturer and operational responsibility is ambiguous. The development of international safety standards (extensions to ISO/TS 15066) is not expected to catch up until 2028 or later.
The Problem of "Specialized Robots Other Than Humanoids Are Sufficient"
Logistics: Specialized robots such as Agility Digit and Amazon Proteus are more efficient.
Elderly Care: Care lifts and transfer assistance robots are superior in terms of safety and cost.
Food Service: Single-axis robotic arms have lower installation costs.
Humanoids are marketed for their "versatility," but they have a fundamental problem: they often cannot compete with specialized robots in terms of economic rationality.
Sales Hooks from a Business Perspective
From the perspective of hydrogen embrittlement and fatigue evaluation, there are specific areas of potential around humanoids:
Shaft Fatigue Testing of Small, High-Torque Motors: Humanoid joints are subjected to high-frequency, high-load rotational motion. Fatigue failure of motor shafts and stress concentration evaluation of rotor cores are critical. Harmonic drive wave gear systems are also components subjected to hundreds of millions of cycles of stress, and the need for long-term reliability evaluation is structurally growing.
Fatigue Life Evaluation of Precision Bearings: If companies like Minebea, NTN, and NSK anticipate producing hundreds of millions of bearings for humanoid robots annually, the need for fatigue testing equipment for lot quality assurance could increase dramatically.
Fatigue Evaluation of Joint Link Components (Al, Ti, Composite Materials): For lightweight robots like the Optimus class, the fatigue properties of carbon fiber reinforced plastics (CFRP) and titanium alloys are crucial. Hydrogen embrittlement, in particular, is a problem that easily arises with lightweight metals.
Stress Corrosion Cracking of Sensor Housings: For humanoid robots operating outdoors or in humid environments, there is a potential demand for stress corrosion cracking (SCC) testing of sensor housings and connectors.
Screw Components Inside Actuators: Rolling fatigue testing of ball screws and planetary gears becomes more difficult as miniaturization progresses. This is an area where electronic science's compact testing equipment can be valuable.
Specific Target Companies
Harmonic Drive Systems: Ariake Plant (Azumino City), Hotaka Plant (Azumino City), Harmonic Precision (Matsumoto City)
Nabtesco: Tsu Seiki Manufacturing Plant, Tarui Plant (Gifu)
THK: Yamagata No. 1 Plant (Obanazawa), Yamagata No. 2 Plant (Shinjo), Kofu Plant, Gifu Plant
MinebeaMitsumi: Karuizawa, Matsuida, Hamamatsu, Bangkok Plant (Overseas but ordered from Japan)
FANUC: Oshino Plant (Yamanashi)
Yaskawa Electric: Nyuzen Plant (Toyama), Nakama Plant (Fukuoka)
In particular, Harmonic Drive's Ariake Plant is currently undergoing a 10 billion yen investment in a humanoid robot expansion line, and the demand for testing equipment and contract evaluation for quality assurance departments is expected to peak in 2026-2027. This presents an opportunity for Electronic Science's equipment proposals and contract testing services.
Clarifying Investment Stance
Aggressive Positions (Clear Winning Strategy)
Harmonic Drive (6324): If medium-term sales targets are achieved, the stock price is more than double the current price.
MinebeaMitsumi: Solid demand for bearings and small motors.
THK: Dual strength in proprietary humanoid development and linear motion components.
Defensive Positions (Value)
Nabtesco: Solid strength in large joints and precision reducers; undervalued PER.
Sony Group: Motive for image sensors.
Cautionary Positions
Harmonic Drive's Short-Term Valuation: Possibility that expectations are already priced in.
Fanuc's Short-Term Valuation: Top Nikkei Contributors in 2026 also have aspects of market thematic rallying.
Medium- to Long-Term Main Target: The strength of Japanese companies lies in components. As the US and China clash over humanoid robot bodies, Japanese component manufacturers can profit from a "neutral arms dealer" position, selling to both sides. The calm conclusion is that the real focus of Japanese humanoid investment is not the body, but the parts.