The Robots Have Clocked In

When Japan Airlines (JAL) deployed humanoid robots at Tokyo's Haneda Airport in May 2026, the industry's message was hard to miss. This was not a press conference stunt but a three-year operational commitment from a legacy aviation carrier in one of the world's most safety-conscious regulatory environments, tasked with solving a problem that is not going away.

Japan's working-age population is projected to decline by 31% between 2023 and 2060.¹ Haneda handles roughly 85.9 million passengers annually.² JAL operates around 4,000 ground handling workers, and Japan is targeting 60 million inbound tourists by 2030, up from a record 42.7 million in 2025.² The demand is going up. The labor supply is going down. Humanoids may be the answer the industry is reaching for.

JAL partnered with GMO AI & Robotics to begin the trial in May 2026, deploying two Unitree Robotics-based humanoid platforms at approximately US$15,400 per unit.³ The robots are tasked with baggage loading, container transport, and aircraft cabin cleaning; exactly the repetitive, physically demanding work that accelerates labor attrition. Their humanoid form factor was deliberately chosen: airports were built for people, not wheeled machines. Humanoids fit the infrastructure that already exists.

BMW's experience makes the commercial case even clearer. After an 11-month deployment of two Figure AI Figure 02 humanoids at its Spartanburg, South Carolina plant, the robots contributed to producing over 30,000 BMW X3 vehicles, loaded more than 90,000 sheet metal components, and accumulated approximately 1,250 operational hours running 10-hour weekday shifts.⁴

That success prompted BMW to expand its humanoid program to Europe for the first time. In February 2026, the company announced it would deploy AEON, the humanoid developed by Hexagon AB's (Nasdaq Stockholm: HEXA B) robotics division, at Plant Leipzig⁵, with a full-scale pilot targeting summer 2026, covering high-voltage EV battery assembly and component manufacturing.⁶ BMW's head of process management added something investors should note: humanoids could allow BMW to bring work in-house that currently flows to suppliers.⁵ That is not a labor story. That is a margin story.

Amazon is perhaps the furthest along in building what every investor in this space is searching for: a deployment flywheel. Through its majority stake in Agility Robotics, Amazon has deployed Digit, a bipedal humanoid for tote handling, in active fulfillment center operations, including a GXO-operated Spanx facility in Georgia, where units move totes between autonomous mobile robots and conveyors without operator intervention.⁷ In March 2026, Amazon acquired New York-based Fauna Robotics for an undisclosed sum, absorbing the 50-person startup's team into its Personal Robotics Group to build out its consumer humanoid presence.⁸ The strategy is two-pronged: own the industrial robot fleet, and own the home robot platform.

The broader deployment picture confirms this is a trend, not an isolated event. Boston Dynamics' electric Atlas has begun commercial deployments with its entire 2026 production allocation committed to Hyundai and Google DeepMind.⁹ AgiBot produced its 10,000th humanoid in late March 2026, scaling from 1,000 units in 2025 to 10,000 within months.¹⁰ In April 2026, Chinese humanoid robots competed publicly in the Beijing E-Town Half-Marathon, with a humanoid named "Lightning", built by Chinese smartphone maker Honor, completing the 21-kilometer course in 50 minutes and 26 seconds, faster than the human half-marathon world record.¹¹

The Brain Is Getting Smarter (Fast)

Today's deployments run on specialized software trained for specific tasks. Tomorrow's market leaders will be defined by something different: generalist AI models that let a robot learn a new job in hours rather than months. The model race is where the long-term investment thesis is being built.

Vision-Language-Action Models: Connecting Perception to Motion

The architectural breakthrough enabling next-generation humanoid intelligence is the Vision-Language-Action model, or VLA. Rather than hardcoding task-specific routines for every motion, a VLA model combines visual perception, natural language understanding, and motor command output in a single framework. A robot running a VLA receives an instruction in plain English, interprets its environment, and executes, without being pre-programmed for the specific scenario.

NVIDIA's GR00T N1 (March 2025) and N1.5 (May 2025) were among the first open, customizable humanoid foundation models, pairing a Vision-Language Model backbone for reasoning with a Diffusion Transformer action module for motor control.¹² NVIDIA (1.96% Weight in KOID as of 4/30/2026) reports that adding synthetic motion data via its GR00T-Dreams simulation pipeline improved task success rates by approximately 40% compared to using only real-world data.¹² Google DeepMind's Gemini Robotics family, built on Gemini 2.0, adds 3D spatial perception and the ability to generate robot code on the fly; its On-Device release in June 2025 made the model lightweight enough to run locally on the robot itself (a critical step for real-world deployment).

Physical Intelligence (π) is producing the most closely watched series of generalist model releases in the space. Its pi-0 model demonstrated multi-robot, multi-task training at scale. Pi-0.5 (April 2025) showed that a robot could clean an entirely unfamiliar kitchen or bedroom from a single high-level voice command. The April 2026 pi-0.7 raised the bar: the model performed tasks it was never trained on, including operating an air fryer it had seen only twice in its entire training dataset, by synthesizing those fragments with web pretraining into a working understanding of the appliance.¹³ Physical Intelligence is reportedly in discussions for a new fundraising round that would push its valuation toward approximately $11 billion.¹⁴

The most consequential model-layer finding of 2026 may be the EgoScale paper, published in February, which provided the first strong empirical evidence that robotics foundation models follow the same data-driven scaling laws as LLMs, and policy performance improves predictably with pretraining data size.¹⁵ In other words, the same compounding capability curve that made ChatGPT transformational appears to apply to physical AI. The companies accumulating the most data have self-reinforcing advantages.

World Models: Giving Robots an Internal Physics Engine

Running alongside VLA development is the emergence of world models, AI systems that simulate the physical environment and predict how objects and forces interact over time. World models give robots the capacity to reason about a task before committing to physical action.¹⁶

The strategic importance of this is direct: unlike LLMs, which trained on the entire text of the internet, physical AI requires embodied data (recordings of how physical matter actually behaves). That data does not exist online in usable form. World models offer a path to generating it synthetically at scale, describing a scenario, simulating it with physics-aware models, and creating training episodes without a single robot in the room. Approximately $6 billion flowed into six or seven world model companies in Q1 2026 alone.¹⁵ Whoever controls the simulation infrastructure controls the ability to scale robot training without scaling physical hardware fleets in proportion.

The ChatGPT Moment: Where Are We?

Bessemer Venture Partners, writing in May 2026, placed the industry squarely in the 'GPT-2.5 moment for robotics'; capabilities are real, scaling laws are beginning to emerge, but the gap between lab performance and the 99.9% reliability that production deployment demands remains wide.¹⁵

Aron Kisdi, Managing Director of Autodiscovery, a UK-based humanoid robotics company, offered perhaps the most grounded take on this question at our KraneShares webinar earlier this year: "I think the ChatGPT moment will happen when the first company publishes earnings saying, 'This humanoid robot just produced 500 million in revenue last year.' It's going to be a more boring ChatGPT moment than a nice video; it will be an earnings report, but that will be the moment where everyone goes, 'I want that as well.'"

Bank of America projects approximately 90,000 humanoid robot shipments in 2026, rising sharply to 1.2 million by 2030, and a global humanoid robot population of 3 billion by 2060.⁹ Morgan Stanley estimates the total addressable market could reach $5 trillion with as many as one billion humanoids deployed by 2050.¹⁷ Whether those numbers prove conservative or optimistic, the direction is not in question.

Is This EV in 2012? The Supply Chain Question Every Investor Should Be Asking

McKinsey put it bluntly in its April 2026 report: the robotics supply chain is the most underappreciated constraint on humanoid scale, and for companies that move now, it is the most significant opportunity in a generation.

Venture capital funding for robotics surged more than threefold between 2023 and 2025, reaching $40.7 billion annually.²⁰ Governments have declared embodied AI a strategic priority; China alone committed a $138 billion state venture capital guidance fund to AI and robotics, among other high-tech sectors.²⁰ But neither capital nor political will resolves the core constraint: the supplier ecosystem for humanoid-specific components is simply not yet built for high-volume production. Understanding where that ecosystem is ready, where it is bottlenecked, and who is racing to own the platform roles that have not yet been claimed; that is the supply chain question that will define a decade of returns in this space.

Opening the Bill of Materials (BoM)

McKinsey's analysis breaks the humanoid hardware stack into five domains: actuators (40–60% of BoM), sensing and perception systems (10–20%), compute and control platforms (10–15%), structural components (5–10%), and battery modules (5–10%).²⁰ Together, these account for 85–90% of total unit costs. The current BoM ranges from roughly $30,000 to $150,000 per unit, while the widely cited long-term target is under $20,000, suggesting significant cost compression is still required before mass-market demand unlocks.²⁰

What makes the supply chain picture strategically distinctive is the mismatch between where value concentrates and where suppliers are ready for high-volume production. Actuators, by far the largest cost driver and primary performance differentiator, depend on one of the least developed supplier ecosystems in the stack. The opposite is true for compute and batteries, which benefit from deep adjacency to the electric vehicle (EV), semiconductor, and consumer electronics industries, where production infrastructure already operates at scale.

Tesla's Optimus illustrates the engineering intensity involved. The robot integrates 28 joint actuators across its body, plus more than 50 additional actuators in its Gen 3 hands alone.²⁰ Each actuator combines a gearbox (30–50% of actuator cost), motor control electronics, a frameless brushless DC motor, precision mechanical components, and sensors; all packed into a compact, thermally constrained form factor.²⁰ There is currently no equivalent of an 'engine supplier' for humanoid actuators. Every leading project is doing custom development. That is both the bottleneck and the opportunity.

China's Advantage Is Not a Labor Story

The cost asymmetry in humanoid manufacturing is structural, and understanding it correctly matters for how investors should think about KOID's geographic allocation. Building Tesla's Optimus Gen 2 without Chinese suppliers would cost approximately three times as much, the bill of materials surging from roughly $46,000 to $131,000.²⁰ Unitree lists its G1 humanoid at approximately $13,500. Tesla's (1.75% Weight in KOID as of 4/30/2026) Optimus humanoid is expected to reach $20,000 to $30,000 at production scale.²⁰ That gap is not primarily about labor costs. It is about ecosystem depth.

China holds approximately 90% of global permanent magnet processing capacity, 40% of precision bearings, 35% of motors, and 30% of power electronics.²⁰ Many of the most critical humanoid subsystems, including motors, harmonic drives, power electronics, battery systems, and sensors, sit close to China's mature EV value chains, enabling supplier reuse, process transfer, and faster scale-up. China recorded 295,000 new industrial robot installations in 2024 and reached an operational stock of approximately 2.03 million factory robots, both world records that accounted for roughly 54% of global deployments that year.²⁰ That manufacturing base is now being redeployed toward humanoids. On the innovation side, China filed approximately 7,700 humanoid-related patents over the past five years, versus roughly 1,560 in the United States.¹⁷

Rather than a single global winner, the humanoid supply chain is more likely to bifurcate. China reaches hardware scale and early cost compression sooner through manufacturing execution and deployment learning. US and European ecosystems differentiate through frontier AI sophistication, system architecture, and high-assurance, safety-certified deployments.

Where the Bottlenecks Are

McKinsey maps supply chain risk across the full hardware stack and identifies three clusters of high-risk components, the choke points most likely to constrain humanoid scale, which are also the areas representing the most attractive opportunities for suppliers that can move early.

Harmonic and strain-wave drives are the clearest bottleneck in the actuator stack. These compact, high-torque-density gearboxes remain concentrated among a small group of manufacturers: Harmonic Drive (2.31% Weight in KOID as of 4/30/2026), Nabtesco (2.21% Weight in KOID as of 4/30/2026), and emerging Chinese players such as Leaderdrive (1.94% Weight in KOID as of 4/30/2026).²⁰ Production is precision-bound and capital-intensive, requiring dedicated tooling, metrology infrastructure, and long qualification cycles. Unlike electronics, where capacity can be added rapidly, harmonic gearboxes are structurally harder to scale. Planetary roller screws present an even more acute risk, a high-load, high-precision niche with a narrow supplier base (including SKF and select high-precision Asian specialists), long lead times, and limited substitution options as OEMs push toward higher payloads and more dynamic motion profiles.

Force and tactile sensing is the second high-risk cluster. Six-axis force and torque sensors, supplied by a limited number of robotics-focused vendors, including ATI Industrial Automation (owned by Novanta Inc; 1.85% Weight in KOID as of 4/30/2026) and OnRobot, are calibration-intensive and feature limited automation in production.²⁰ Unlike motors or control electronics, these sensors benefit little from automotive or consumer spillover; scaling depends on expanding high-precision calibration infrastructure that remains concentrated. Tactile sensors face a different challenge: many leading solutions originate from startups or research institutions, creating a fragmented, unstandardized landscape with no dominant architecture. Dexterous hands sit at the intersection of both clusters.

As Aron Kisdi explained during our webinar, "When you grab a glass of water, you don't think about it, but your skin is detecting slip, moisture, and texture simultaneously. To implement that, the hand itself needs to sense all of those things. There is a lot of research still needed." McKinsey's analysis confirms the structural point: hands combine compact actuation and robotics-grade sensing, both still lacking standardized architectures and fully scaled supplier ecosystems, making them a potential gating factor and a key opportunity when deployment scales.

The Platform Land Grab Is Already Underway

The most important near-term signal in the supply chain story is not which components are scarce; it is who is positioning now for the platform roles that will define the ecosystem once humanoid volumes reach industrial scale.

In actuation, Schaeffler (1.93% Weight in KOID as of 4/30/2026) signed three humanoid actuator partnerships in five months: Neura Robotics, UK-based Humanoid, and China's Leju Robotics, codeveloping next-generation strain-wave gear actuators across wheeled and bipedal platforms.²⁰ The company expects up to 10% of group sales in 2035 to come from new sectors, including humanoid robotics. Bosch entered into a partnership with Neura Robotics covering component supply and motor production, while its Boyuan Capital arm formed a joint venture with Chinese humanoid developer Galbot.²⁰ Magna (1.93% Weight in KOID as of 4/30/2026) took an equity stake in Sanctuary AI. The pattern is consistent: these companies are codeveloping humanoid-specific subsystems rather than adapting off-the-shelf components, and design wins at the prototype stage translate into production incumbency once architectures stabilize.

In compute, Qualcomm launched the Dragonwing IQ10, a humanoid-specific processor, working with Figure AI and Neura Robotics to define next-generation compute architectures.²⁰ Infineon (2.59% Weight in KOID as of 4/30/2026), NXP Semiconductors NV (2.54% Weight in KOID as of 4/30/2026), STMicroelectronics NV (2.75% Weight in KOID as of 4/30/2026), and Texas Instruments Inc (2.46% Weight in KOID as of 4/30/2026), all formalized humanoid-specific product integrations covering motor control, real-time communications, sensor fusion, and safety logic.²⁰ In manufacturing, Jabil Inc (2.21% Weight in KOID as of 4/30/2026) became the worldwide production partner for Apptronik's Apollo humanoids.²⁰ SoftBank announced a $5.4 billion acquisition of ABB's robotics division, explicitly framing physical AI as its 'next frontier.'²⁰

The inflection point that triggers full industrialization, when predictable multiyear volumes justify dedicated tooling, certification investment, and platform-level R&D, has not arrived yet. But it may arrive sooner than expected. The companies defining standardized interfaces and subsystem form factors before that moment are the ones that will capture long-term ecosystem leverage. That is the CATL lesson applied to the humanoid industry today.

What to Watch: Near-Term Catalysts for H2 2026

Macro theses are useful. Catalysts are actionable. Here is the roadmap of events most likely to move the conversation and the market over the next six months.

Tesla Optimus Gen 3: High Stakes, Moving Timeline

Tesla announced in January 2026 that it would convert its Fremont Model S/X assembly line into the first-generation Optimus production line, designed for capacity of up to one million units per year.²⁴ On the Q1 2026 earnings call, Elon Musk pushed the production start to late July or August 2026, after earlier acknowledging that Optimus units in Tesla's facilities were "still very much in the R&D phase," operating primarily for data collection rather than productive manufacturing tasks.²¹ External commercial sales are not expected until late 2026 at the earliest; consumer availability targets late 2027.²¹

Tesla's timeline has shifted multiple times and should be treated as directionally correct, but exact timing is uncertain. The more durable point: Tesla's vertical integration of actuator design, compute, and AI training infrastructure creates potential cost advantages of 30% to 40% over competitors dependent on third-party suppliers.¹⁸ We believe that when the world's highest-profile EV manufacturer reallocates factory space from vehicles to humanoid production, it signals the sector's trajectory regardless of exact delivery dates.

Unitree IPO: The Sector's First Profitability Proof Point

Unitree Robotics filed for a 4.2 billion yuan (~$610 million) IPO on Shanghai's STAR Market in March 2026, with CITIC Securities as lead underwriter and a target valuation of approximately $7 billion.²² This is the most significant near-term liquidity event in the humanoid robotics sector, and the financials may surprise people who assume the industry is purely cash-burning.

Gross margins across Unitree's humanoid and quadruped segments reached approximately 60% in 2025.²² Revenue from humanoid robots surpassed quadruped robots for the first time last year, accounting for over 51% of total revenue.²² The company shipped over 5,500 humanoid robots in 2025, surpassing the combined output of all US competitors, including Tesla, Figure AI, and Agility Robotics, and is targeting 20,000 units in 2026.²³

The pricing trajectory is equally important. Average humanoid robot prices fell from approximately $85,000 in 2023 to $25,000 in 2025, even as gross margins improved.²² That combination of volume growth, rapid price compression, and expanding margins is the fingerprint of a maturing manufacturing platform. Unitree's entry-level R1 model now starts at $5,900.²³ If the IPO succeeds, it could reset industry valuations globally and improve financing conditions across the entire supply chain.

Conclusion: Physical AI Is Emerging: Get Ready

Humanoid robotics is completing the transition from science project to investment thesis. The evidence is no longer theoretical. Japan Airlines is deploying robots at Haneda Airport. BMW validated Figure 02 through 1,250 operational hours on an active automotive line and is scaling to Leipzig. Amazon is building a consumer physical AI platform through Fauna. Unitree is filing for a $7 billion IPO with 60% gross margins. Physical Intelligence's pi-0.7 is exhibiting the first signs of compositional generalization. The capital is committed: over $34 billion into robotics in 2025 alone.¹⁶

This is not the ChatGPT moment yet. That requires a robot performing a complex task in an unfamiliar environment, without human supervision, reliably enough for mass commercial deployment. But we are building toward it, and the scaling laws that drove LLM capability improvements appear to apply here, too. The companies accumulating the most embodied training data through deployed fleets have self-reinforcing advantages that will compound.

The supply chain question deserves particular attention. The EV analogy from 2012 suggests that the most durable value will accrue to the component categories with the highest technical barriers: actuators and reducers, force and torque sensors, and dexterous end effectors. The CATL lesson is clear: vertically integrated component manufacturers that achieve scale in a technically complex category at the right moment in an industry's development curve can build positions extraordinarily difficult to displace.

KOID holds 50 companies across the full humanoid and physical AI ecosystem: the semiconductors and AI software that power the brain, the actuators, sensors, materials, and precision control systems that form the body, and the systems integrators and humanoid OEMs that assemble the whole. Equal-weighted to avoid concentration in large-cap US technology names that would dominate a market-cap-weighted approach, and global representation across the US, China, Japan, and Europe, KOID is designed to capture the potential opportunity across the full value chain as the physical AI era scales.

Similar to what NVIDIA and semiconductors did for digital AI, these are the picks and shovels for physical AI. The play is emerging. Get ready.

---

Holdings are subject to change. All company weight %'s in KOID are from Bloomberg as of 4/30/2026.

For KOID standard performance, top 10 holdings, risks, and other fund information, please click here.

Citation:

1. "Japan's Working-Age Population to Fall 31% by 2060." OECD Employment Outlook, 2024.
2. "Tourism Innovation in Japan's Haneda Airport Integrates Humanoid Robots for Baggage Handling." Travel And Tour World, April 2026.
3. "Japan Airlines Begins Humanoid Robot Trials at Tokyo's Haneda Airport as Labor Shortages Bite." CNBC, 5/1/2026.
4. "F.02 Contributed to the Production of 30,000 Cars at BMW." Figure AI Blog, 11/19/2025.
5. "BMW Group to Deploy Humanoid Robots in Production in Germany for the First Time." BMW Group Press Release, 2/27/2026.
6. "Inside BMW's Humanoid Robot Reveal: AEON at Plant Leipzig." BMW Blog, 3/3/2026.
7. "Humanoid Robots in Industrial Manufacturing: What They Can (and Can't) Do in 2026." EVS International, April 2026.
8. "Amazon Acquires Fauna Robotics as It Expands Into Home Humanoid Robots." Robotics and Automation News, 3/26/2026.
9. "May 2026: Humanoid Industry Accelerates with Real Deployments, Beijing Robot Marathon Success, and Production Scaling." Humanoid.press, May 2026.
10. "China's Humanoid Robot Output to Surge 94% in 2026; Unitree and AgiBot to Capture Nearly 80% Market Share." TrendForce, April 2026.
11. "Beijing E-Town Half-Marathon: Chinese Humanoid Robots Compete." Humanoid.press, April 2026.
12. Bjorck, J. et al. "GR00T N1: An Open Foundation Model for Generalist Humanoid Robots." NVIDIA, arXiv:2503.14734, March 2025.
13. Loizos, C. "Physical Intelligence Says Its New Robot Brain Can Figure Out Tasks It Was Never Taught." TechCrunch, 4/16/2026.
14. Loizos, C. "A Peek Inside Physical Intelligence, the Startup Building Silicon Valley's Buzziest Robot Brains." TechCrunch, 1/30/2026.
15. "Bessemer Predicts: Robotics and Physical AI." Bessemer Venture Partners, May 2026.
16. "How Physical World AI Could Reshape Our Economy." Generation Investment Management, March 2026.
17. "Humanoids: 1bn Robots and $5tn Revenues by 2050, China Is in Pole Position." Morgan Stanley Research, 4/28/2025.
18. "Understand the Industrial Chain in 'One Page': Humanoid Robots." 36Kr English, September 2025.
19. "The Global Humanoid Robot Supply Chain: Architecture, Economics, and Strategic Implications." Gerra Research, 2025.
20. "Turning Humanoid Supply Chain Constraints into Billion-Dollar Wins." McKinsey & Company, April 2026.
21. Tesla Q1 2026 Update (assets-ir.tesla.com).
22. "China Robot Maker Unitree Files for $610 Million Shanghai IPO." Rest of World, 3/31/2026.
23. "Beyond Kung Fu: Unitree Launching Humanoid Robot for Household Chores and Elder Care." Notebookcheck, 3/20/2026.
24. Electrek's Q1 2026 earnings coverage.