Chaos and Order

💡 왼쪽 원문을 읽으면서 오른쪽에 따라 써보세요. Tab 키로 힌트를 받을 수 있습니다.

원문 렌더가 준비되기 전까지 텍스트 가이드로 표시합니다.

Introduction — Why Humanoids, Why Now

One of the dominant talking points in the first half of 2026 is, without question, humanoid robots. The bipedal, human-shaped machine that was dismissed as "science fiction" only a few years ago now appears in demo videos carrying parts along automotive assembly lines and moving boxes in logistics warehouses. A machine that was once a curious laboratory experiment is increasingly being named in the market under a new grammar: that of an "investable asset."

Three forces are said to be converging behind this shift. First, the progression from large language models (LLMs) toward large action models (LAMs) and vision-language-action (VLA) models that handle behavior, not just text. Second, the rising performance of edge AI chips that let robots reason on-device rather than depending on the cloud. Third, the gradual decline in hardware costs for actuators, sensors, and batteries.

> This article is for informational and educational purposes only and is not investment advice or a solicitation. All investment decisions and the responsibility for them rest entirely with you, and you should consult a qualified professional where appropriate. The company names below are subjects of fact-based analysis, not buy or sell recommendations for any specific security.

In this article we cover the state of humanoid technology, the key companies, use-case scenarios, the value chain, and both the bull and bear perspectives in a balanced way. In particular, we try to look past the flashy demo videos toward the fundamentals: actual contracts, deployed units, unit costs, and cash flow, the "boring numbers" that matter.

1. State of the Technology — LAM, VLA, and Edge AI

1.1 What Has Changed

The industrial robots of the past were "programmed arms" repeating fixed motions. If a part was not placed at exactly the right spot on the conveyor, the machine would stop or perform the wrong action. By contrast, recent humanoids aim for "generalization": understanding a scene seen through a camera, taking natural-language instructions, and picking up an object they have never encountered before.

The key enabler is the vision-language-action (VLA) model. Unlike an LLM that handles only text, a VLA links camera input (vision), human instruction (language), and the robot's joint-control signals (action) within a single model. In other words, it is an attempt to have one neural network hear the sentence "put the red cup in the sink," locate the red cup on screen, and simultaneously control dozens of joints in the arm and fingers.

[Traditional Industrial Robot] [Humanoid + VLA/LAM]

Repeat fixed coordinates Understand a scene visually

Pre-programmed Interpret natural language

Fragile to change Generalize to new objects

Single-task specialized Transfer across many tasks

Fixed installation Free to walk and move

1.2 The Difference Between LAM and VLA

The terms are often used interchangeably, so a clarification helps. VLA tends to emphasize a single "see-hear-move" policy model, while LAM is often used as a broader concept for "a large model that outputs actions." In practice, both can be understood as models extended from "predict the next text token" to "predict the next action."

| Aspect | LLM | VLA / LAM |

| --- | --- | --- |

| Input | Text | Images, text, sensors |

| Output | Text | Joint torque, motion sequences |

| Training data | Web text | Teleoperation, simulation, video |

| Metric | Accuracy, benchmarks | Task success rate, safety |

| Bottleneck | Compute | Physical action data |

1.3 Why Edge AI Matters

For a robot to move safely in the same space as people, latency is decisive. If you handle the instant before a fall via a cloud round trip, it is already too late. That is why edge AI, processing a large share of inference on the robot's own chip, is cited as essential. Nvidia is said to be targeting this area through its robot computing platforms, such as the Jetson family and the Isaac robotics stack.

Edge inference is not only about speed. Constantly streaming what a camera sees in a home or factory to an external server is costly in terms of communication, security, and privacy. That is why an "on-device" design that completes core inference inside the robot is frequently cited as a precondition for commercialization.

1.4 The Data Bottleneck, Teleoperation, and Simulation

Interestingly, many analyses point to "data" as the biggest bottleneck for robot AI. Text is abundant on the internet, but physical action data, such as "the motion of a hand grasping an object," is scarce. So teleoperation, where a person directly controls the robot to gather data, and large-scale synthesis in simulation are pursued in parallel.

Paths to acquiring physical action data

[Teleoperation] Human pilots robot via VR -> high quality, costly

[Simulation] Mass generation in virtual env -> cheap, sim-to-real gap

[Real-world] Deployed robots collect directly -> scale, safety burden

Teleoperation yields high quality but is expensive because it consumes human time, while simulation can generate data cheaply at scale but leaves the homework of the sim-to-real gap. Ultimately, many see the core of the competition as who first builds the virtuous loop in which deployed robots themselves gather data and learn again. For this reason, the "data flywheel" is one of the most frequently cited keywords in the humanoid theme.

2. Key Companies — Figure, Tesla, 1X, and Others

Each company's approach differs slightly. The table below is an overview based on publicly available information; detailed specifications may change over time.

| --- | --- | --- | --- |

2.1 Figure AI

Figure is a U.S. startup focused on pure humanoid development, putting its in-house AI model Helix front and center. It drew attention with reports of a pilot at a BMW factory and is said to have the home market in its sights as well. Bulls view its vertically integrated strategy, not relying on external models, as a differentiator.

2.2 Tesla Optimus

Tesla emphasizes that the vision AI and mass-production know-how accumulated in self-driving (FSD) can be transplanted into Optimus. The reported scenario is to deploy in its own plants first to gather data and lower costs, then expand outward. That said, the fact that production timelines and target costs have been revised several times is also used as a basis for caution.

2.3 1X Technologies

1X emphasizes home-environment safety as a differentiator with its NEO model. A design that lowers the risk of contact with people through a soft exterior and lightweight structure, rather than a rigid metal exoskeleton, is often cited as a feature. There were reports that OpenAI participated in its funding, and the trend is to emphasize software and safety design together.

2.4 Boston Dynamics, Agility, and Unitree

Boston Dynamics has demonstrated high-end mobility with its new electric Atlas and is of particular interest to Korean investors because it sits under Hyundai Motor Group. Agility Robotics' Digit, specialized for warehouses, has seen multiple reported pilots, and China's Unitree is assessed as targeting the entry-level market with relatively low prices.

2.5 One Company, Two Views

Bulls argue that Tesla's manufacturing capacity, Figure's vertical integration, and 1X's safety design can each become a moat. Skeptics, by contrast, point to the gap between demo videos and actual mass production and monetization. The demos in controlled environments are impressive, but coping with the infinite variables of a home or a general workplace is an entirely different problem.

3. Use-Case Scenarios — Cobots, Logistics, and Housework

The prevailing view is that humanoid adoption will spread in stages.

Adoption difficulty (low -> high)

[Stage 1] Collaborative robots (cobots) — fixed support tasks in factories

[Stage 2] Logistics — sorting, stacking, transport in warehouses

[Stage 3] Services — assistance in stores, hospitals, facilities

[Stage 4] Housework — cleaning, tidying, care at home (hardest)

3.1 Why Factories and Logistics Come First

Factories and warehouses have relatively controlled environments, repetitive tasks, and an easy-to-calculate return on investment (ROI). Above all, these are areas suffering severe labor shortages, so the incentive to adopt is strong. That is why many analyses expect early commercialization to concentrate on B2B industrial sites.

Logistics in particular features many relatively standardized "pick and place" tasks, so it is often cited as the first proving ground for humanoids. The argument is that adoption gains its first justification in areas where people are hard to find, such as night shifts and dangerous or repetitive work.

3.2 The Potential of Service Areas

Service areas such as store guidance, hospital supply transport, and facility inspection have more variables than factories but are more controllable than homes, so they are cited as an intermediate stage. The catch is that as direct interaction with people increases, so do safety and legal-liability issues.

3.3 Why Housework Is Hardest

By contrast, the home is an environment that is almost impossible to standardize. Every house has a different layout, pets and children move around, and safety requirements are extremely high. Therefore the mass adoption of a "robot that does chores" is cited as the last stage. Flashy home demos are frequently released, but the common cautious view is that significant time is needed before real mass adoption.

3.4 Comparing Adoption Conditions by Stage

Organizing the conditions each stage requires into a table makes the priority clearer. You can confirm that the more controllable the environment and the easier the ROI calculation, the earlier adoption comes.

| --- | --- | --- | --- | --- |

This table does not assert an absolute order but should be read as a conceptual summary explaining "why industrial sites are cited first." From an investment standpoint, it helps to examine which stage each company makes its focus, and what the entry barrier of that stage is.

4. The Debate Over Price Decline and Adoption Timing

From an investment standpoint, the hottest issue is "when and at what price will it spread."

4.1 The Price Curve

Some manufacturers were reported to have presented a goal of lowering the per-unit cost to roughly USD 20,000 to 30,000 over the long term. That is close to the price of a car. However, this is a future target premised on mass production, and current prototype and small-batch costs are known to be far higher.

Price-decline scenario (conceptual, unit: USD 10,000)

Per-unit cost

20 |#

16 |# #

12 |# # #

8 |# # # #

4 |# # # # # #

+--------------------

Early -> Mass production -> Maturity

4.2 Adoption-Timing Scenarios

Views also diverge on adoption timing. The following organizes frequently cited rough scenarios; it does not assert a specific date.

| Segment | Optimistic scenario | Cautious scenario |

| --- | --- | --- |

| Full industrial adoption | Meaningful units within years | Validation delays, gradual spread |

| Service-area spread | Right after industrial adoption | Held back by safety and regulation |

| Start of home adoption | Cited at a relatively early point | Pushed well into the future |

| Per-unit cost at car level | Quickly via economies of scale | Slowly due to precision parts |

4.3 Optimists and the Cautious

Optimists expect that, as seen with batteries and electric vehicles, economies of scale and learning effects will drive costs down quickly. The cautious, by contrast, warn that precision actuators and high-quality sensors, unlike simple parts, see slower cost declines, and that "production hell" should not be underestimated. Indeed, the auto industry is cited as a counterexample, having needed long timelines and enormous capital to stabilize mass production.

5. The Value Chain — Where the Real Beneficiaries Are

The analogy that the people who sold pickaxes during the gold rush were the ones who made money is frequently cited. The view in the humanoid theme is that one should look not only at finished-product makers but at the entire parts and materials supply chain.

| Area | Key parts | Watch points |

| --- | --- | --- |

| Actuators | Precision motors, reducers, ball screws | Large share of cost, precision is key |

| Precision motors | Frameless motors, BLDC | Torque density and heat management |

| Reducers | Harmonic and cycloidal | Precision and durability, concentrated supply |

| Sensors | Cameras, torque sensors, tactile sensors | Foundation of perception and safety |

| Batteries | High-density cells, power management | Balance of runtime and weight |

| Semiconductors | Edge AI chips, GPUs | On-device inference performance |

| Materials | Lightweight alloys, artificial skin | Weight, durability, safety |

5.1 Actuators and Reducers

A large share of a humanoid's cost goes to the actuators that move the joints and the reducers that precisely step down the motor's rotation to amplify force. Reducers in particular are demanding in precision and durability, so some analyses note their supply is concentrated among a few firms. Bulls see this area as "the bottleneck and therefore the beneficiary," while bears see slow cost declines here potentially holding back adoption.

5.2 Sensors, Batteries, Semiconductors, and Materials

Sensors are the foundation of perception and safety. Not just cameras but torque sensors that detect force and fingertip tactile sensors are cited as keys to delicate manipulation. Batteries hinge on the balance between runtime and weight, and semiconductors determine on-device inference performance. Materials such as lightweight alloys and artificial skin must satisfy weight, durability, and safety at the same time.

Value-chain flow

Materials/parts -> Subsystems -> Finished robot -> Software/services

(motors,sensors) (joints,hands) (Figure, etc.) (VLA models, ops)

Bulls emphasize that, whichever finished-product maker wins, the parts supply chain benefits in common. Bears, by contrast, note that the market itself is still small, so the earnings contribution of supply-chain firms may be limited for the time being.

5.3 A Conceptual Cost Breakdown

Public information on where the cost of a single humanoid concentrates is limited, but the drivetrain, such as actuators and reducers, is generally cited as taking a large share. The following is a conceptual distribution, not exact figures.

Cost-structure concept (not measured, for a sense of proportion)

Drivetrain (motors/reducers) ######## large

Sensors/vision #### medium

Compute/semiconductors ### medium

Battery/power ## small

Structure/materials ## small

What this distribution implies is that the key to lowering unit cost lies in the drivetrain. Many analyses hold that the mass-production cost of motors and reducers must fall for the overall unit cost to come down meaningfully. Therefore, when looking at the value chain, tracking "which part drives the cost decline" is cited as a useful perspective.

6. Market-Size Outlook — How Far Can We Trust the Numbers

Various institutions and companies have presented very large long-term potential for the humanoid market. Some forecasts were reported to cite a scale of trillions of dollars decades into the future. However, such long-term estimates are highly sensitive to assumptions; changing a single variable such as adoption rate, unit cost, or replacement cycle can drastically change the result.

Sensitivity of market-size estimates (conceptual)

Optimistic ───────────────> Very large market

Neutral ──────> Large market

Conservative ──> Meaningful but limited

* Even the same model yields very different results by input assumptions

Accordingly, much advice holds that, rather than the flashy total addressable market (TAM) figure itself, the more trustworthy posture is to ask "from what assumptions did this number come." A long-term TAM is only a directional reference, not a figure that guarantees revenue for any particular year.

7. Korean Companies in Focus

The humanoid theme is also of direct interest to Korean investors.

7.1 Hyundai Motor Group and Boston Dynamics

Hyundai Motor Group is well known for having acquired Boston Dynamics, and demonstrations of high-end mobility via the electric Atlas have been released steadily. While some see long-term synergy in combining auto-manufacturing capability with robot technology, cautious views also exist regarding the timing of commercialization and monetization.

7.2 Samsung, LG, and the Parts Supply Chain

Large conglomerates such as Samsung and LG were reported to have shown interest in the robotics field. Beyond finished humanoids themselves, it is frequently noted that Korean companies are positioned throughout the value chain in batteries, semiconductors, precision motors, reducers, and sensors. That said, much advice holds that, in phases where share prices move on theme expectations alone, it is important to distinguish actual revenue contribution from expectation.

8. Safety, Standards, and Regulation — The Hidden Variables

For a humanoid to work in the same space as people, technology alone is not enough. Many analyses hold that an "invisible infrastructure" of safety standards, certification, and liability rules must also be in place for adoption to proceed.

8.1 Safety Certification and Standardization

Industrial robots already have standards such as safety fences, emergency stops, and collision detection. But a humanoid that moves freely beside people has much that existing standards do not cover. Items such as force limits, behavior on collision, and stopping in unpredictable situations are reportedly being newly discussed. There is a view that adoption speed may be limited until standardization takes hold.

8.2 Liability and Insurance

When a robot causes an accident, who is responsible: the manufacturer, the operator, or the software provider? Social consensus on this question is cited as being at an early stage. The more insurance and legal frameworks are established, the lower the adoption burden on companies, but the process itself takes time, which is a variable.

The three axes governing adoption

Tech maturity ─┐

Cost decline ─┼─▶ Actual adoption speed

Rules/safety ─┘

* Even if one axis is fast, the whole converges to the slowest axis

8.3 Jobs and Social Acceptance

A humanoid simultaneously fills labor shortages and displaces jobs. If social acceptance is low, regulation may tighten and slow adoption; conversely, if labor shortages are severe, adoption may accelerate. From an investment standpoint, there is advice to also examine the direction of this social variable.

9. Multiple Perspectives — Bull vs. Bear

8.1 The Bull Case

- Global population aging and labor shortages create structural demand.

- VLA/LAM and edge AI advances are rapidly improving generality.

- Tech giants including Nvidia are pouring in platforms and capital.

- Mass-production capability proven in autos and EVs can be transplanted.

- The "data flywheel," where deployed robots gather data, is beginning to work.

8.2 The Bear Case

- The gap between demos and real commercial reliability remains large.

- Precision-part cost declines may not be as fast as hoped.

- Social consensus on safety, regulation, and liability is immature.

- Current valuations may have priced in future expectations excessively.

- There is a track record of production timelines and target costs being repeatedly pushed back.

Same facts, different interpretations

[Fact] Flashy demo videos have increased

├─ Bull: technology is maturing fast

└─ Bear: demos and mass production are different problems

[Fact] Giants are pouring in capital

├─ Bull: the ecosystem is thickening

└─ Bear: expectations may spill into overheating

10. Risks and Checkpoints

The following organizes items worth reviewing before any investment decision.

1. **Overheating risk** — The hotter the theme, the more expectations are priced into shares. If results miss expectations, the correction can be sharp.

2. **Uncertain commercialization timing** — Keep in mind the possibility that "it will be ready next year" promises get repeatedly delayed.

3. **Cash burn** — Enormous R&D and capital expenditure are needed up to mass production, and cash burn is heavy before monetization.

4. **Intensifying competition** — Price competition with low-cost latecomers (especially Chinese firms) can pressure margins.

5. **Regulation and ethics** — Job displacement, safety incidents, and liability can act as policy variables.

6. **Concentration and theme risk** — Excessive exposure to a single theme raises volatility. The principle of diversification applies here too.

> Checkpoint: Rather than demo videos, tracking (1) actual paid contracts and deployed units, (2) uptime and failure rates, (3) the trajectory of unit costs, and (4) cash flow is cited as a more trustworthy set of signals.

11. An Investor's Self-Check Checklist

It can help to answer the following questions for yourself. The right answers differ from person to person.

- How many years is my time horizon for this theme.

- Am I distinguishing flashy demo videos from actual revenue and contracts.

- Am I exposed to finished products or to the value chain (parts and materials).

- Can I state the core logic of the bear case in one sentence.

- Is this position's share of my total assets appropriate.

- Have I set the range of loss I can tolerate if expectations are wrong.

- Is my information from primary sources (filings and results) or from mood and rumor.

Self-check flow

Review thesis -> Review sizing -> Review scenarios -> Exit plan

(why buy) (how much) (if wrong?) (when to trim)

12. FAQ

**Q. Is the humanoid robot business making money right now?**

Most pure humanoid companies are said to be at a pre-large-scale-monetization stage. Demos and pilots are active, but the common analysis is that meaningful revenue and profit are tasks for the future.

**Q. Which is more advantageous, finished-product firms or parts firms?**

It is hard to assert that one side is absolutely advantageous. Finished products have large upside if they succeed but also high uncertainty, while parts and materials are relatively stable but dependent on market size. The right answer depends on your own risk tolerance.

**Q. Is the low-cost push by Chinese firms a threat?**

Latecomers with strong price competitiveness are cited as a margin-pressure factor. However, there is a counterargument that non-price variables such as safety, reliability, and the software ecosystem also matter.

**Q. Is now the time to enter?**

This article does not recommend buying or selling at any specific time. Timing decisions should be made through your own goals, horizon, and risk tolerance, and consultation with a professional.

13. Conclusion

Humanoid robots are a symbol of the larger trend of "Physical AI." As a turning point at which AI, once confined to digital space, gains hands and feet in the physical world, the long-term potential is certainly judged to be large.

That said, the size of the potential and the realization of investment returns are separate matters. It is important to distinguish theme excitement from actual fundamentals, review both bull and bear logic, and make a judgment suited to your time horizon and risk tolerance. In the end, it is worth remembering that the trustworthy signals lie not in flashy videos but in the boring numbers of contracts, deployments, unit costs, and cash flow.

> Once again, this article is for informational and educational purposes only and is not investment advice or a solicitation. All investment decisions and the responsibility for them rest with you, and you should consult a qualified professional where appropriate.

References

- Reuters, robotics and humanoid industry trends — [reuters.com](https://www.reuters.com/technology/)

- Bloomberg, humanoid robots and AI capital flows — [bloomberg.com](https://www.bloomberg.com/technology)

- CNBC, Tesla Optimus and robotics coverage — [cnbc.com](https://www.cnbc.com/technology/)

- Financial Times, AI and robotics investment analysis — [ft.com](https://www.ft.com/artificial-intelligence)

- Nvidia, robotics (Isaac) platform — [nvidia.com](https://www.nvidia.com/en-us/industries/robotics/)

- Figure AI official site — [figure.ai](https://www.figure.ai/)

- 1X Technologies official site — [1x.tech](https://www.1x.tech/)

- Boston Dynamics official site — [bostondynamics.com](https://bostondynamics.com/)

- Yahoo Finance, robotics-related quotes — [finance.yahoo.com](https://finance.yahoo.com/)