Humanoid Robotics in 2026: Where the Real Investment Bets Are
Global humanoid robotics investment is accelerating across every major market simultaneously. China invested $6.95 billion in Q2 2026 alone, six times the year-prior figure. Korea closed its largest-ever humanoid round. Norway's 1X released the most advanced commercial robot hands to date. Toyota spun out a $1.1 billion robotics company in six months. The US data layer is consolidating around NVIDIA and open-source frameworks. And Anthropic published research this week showing Claude controlling physical robots in real environments. Here is what is actually happening across the stack right now.
Capital Is Moving in Every Direction
Three rounds this week tell the story:
China: LimX Dynamics raised $200 million pre-IPO at a $2.21 billion valuation, with a Hong Kong listing in progress. Unitree has been fast-tracked for a Shanghai STAR board listing at a ~$5.9 billion valuation, with revenue up 335% year-on-year after shipping approximately 5,500 units in 2025. China now has more than 100 active humanoid companies. LimX founder Will Zhang cited the Chinese EV listings of Nio, Xpeng and Li Auto as the template: raise, list, or risk disappearing.
Korea: Holiday Robotics closed a $105 million Series A, the largest Korean humanoid round to date, on the thesis that hand dexterity is the primary determinant of commercial utility.
Japan/US: Walden Robotics raised $300 million at a $1.1 billion valuation six months after spinning out of Toyota, signalling how seriously automotive OEMs are treating the question of whether to buy from the robotics supply chain or build inside it.
One structural factor cuts across all of this: China holds approximately 90% of global permanent magnet processing capacity, 40% of precision bearings, 35% of motors, and 30% of power electronics, making the bill of materials for a competitive humanoid roughly three times more expensive to build without Chinese suppliers. LimX founder Zhang said the technology has crossed the "0 to 1" threshold, and the next barrier is product-market fit, not proof of concept.
Hardware: Hands Are Still the Bottleneck
Two different philosophies on the hands problem, both funded heavily this week:
1X Technologies' NEO hands: 25 actuated degrees of freedom, tendon-driven, gear ratios of 5:1 to 15:1 (vs industry standard 100:1 to 200:1), trading peak torque for speed and compliance. Every joint is simultaneously a motor and a sensor, with high-resolution tactile sensing built in.
Holiday Robotics' Friday: 64 total degrees of freedom, with 40 dedicated to the hands alone. More DOF, different bet on what commercial dexterity requires.
Every actuated joint in a humanoid runs on bearings. Goldman Sachs revised its 2035 humanoid market forecast sixfold to $38 billion, and Morgan Stanley is projecting approximately 300x growth in the robot bearings market through 2050, reflecting the minimum mechanical infrastructure required to support deployments at scale. The companies supplying this layer are mostly industrial manufacturers, not startups.
Data: Real-World Deployment Is the Moat
NVIDIA integrated Isaac GR00T 1.7 into Hugging Face's LeRobot framework, combining its simulation infrastructure with the open-source robotics dataset ecosystem. The combined platform now encompasses over 350,000 robot trajectories and 57 million datapoints. NVIDIA's own research shows adding synthetic motion data through GR00T-Dreams improved task success rates by approximately 40% compared to real-world data alone.
Simulation helps, but real-world manipulation data is where the moat is. Figure AI's Figure 02 produced over 30,000 vehicles at BMW's Spartanburg facility, handling more than 90,000 sheet metal components across approximately 1,250 operational hours, a data collection event at commercial scale, not just a production milestone. Companies running commercial deployments in 2026 and 2027 are building datasets that later entrants cannot replicate through simulation.
Anthropic: Claude Is Now Controlling Robots
Anthropic published research this week showing Claude can control physical robots directly, demonstrated on a real Unitree Go2 platform alongside quadrupeds and robotic arms. Direct joint control still fails in most cases, as the model lacks the fine-grained motor precision to replace purpose-built controllers. Where it works is as a supervisory layer: Claude oversees a pretrained controller, interprets the environment through orientation tools, and handles navigation and manipulation tasks that require reasoning rather than precise actuation.
Foundation models handling the decision layer while dedicated controllers handle execution is where the architecture is heading. Anthropic is building physical AI capability while simultaneously committing $21.6 billion to Australian data centre infrastructure, and the AI companies approaching public markets in 2026 are not positioning as pure software businesses.
The Investment Read
Unitree's G1 is now $13,500, down from $90,000 a year ago. Hardware is commoditising faster than most 2024 forecasts anticipated, and Chinese manufacturers have a structural cost advantage that compounds with scale. The companies competing on volume in that market will face thin margins.
The constraint layer is where the durable positions sit:
Actuator designers
Tactile sensor manufacturers
Tendon system engineers
Precision bearing suppliers
Real-world deployment operators building training datasets at commercial scale
The EV transition is the useful reference: vehicle OEMs got the attention, battery chemistry companies and power management chip designers captured the margins. Humanoid robotics is at that equivalent moment, and most of the private companies in the constraint layer are not yet on most investors' radars.
NonPublic Pty Ltd (ABN 49 607 216 928) holds Australian Financial Services Licence #482668. Investments are available to wholesale and sophisticated investors as defined under the Corporations Act 2001. This content is general in nature and does not constitute financial product advice. It does not take into account your objectives, financial situation, or needs. Investing in private markets involves significant risk, including the potential loss of your entire investment. Past performance is not a reliable indicator of future results. You should obtain independent financial advice before making any investment decision.
