Global humanoid robotics and embodied AI commercialization Report Interpretation

Overview

This global humanoid-robotics update examines whether recent technical progress is translating into usable, productive deployment. Morgan Stanley sees evidence of broadening commercialization and stronger investment in embodied-AI models and data, while stressing that real-world task performance remains the industry’s critical hurdle.

Core views

Morgan Stanley frames the sector through Moravec’s Paradox: locomotion has improved materially, but perception, reasoning and dexterous manipulation in changing physical environments remain much harder. The World Humanoid Robot Games illustrate the change in focus. The 2026 event included 666 teams and 2,056 robots, versus roughly 280 teams in 2025, and Tiangong ran 100 metres in 8.64 seconds versus 21.5 seconds in 2025. Yet the report places greater importance on the Games’ shift toward 21 real-world, multi-step scenarios across factories, homes, hospitality, retail, hospitals and emergency response. Fully autonomous operation carried twice the scoring weight of remote operation in relevant scenarios, while new dexterous-hand contests tested fine manipulation. Morgan Stanley’s conclusion is that the relevant benchmark has moved from predefined motion to useful, adaptable work at productive efficiency. Commercial activity is expanding, although deployments remain early. Global shipments reached 19,000 units in 1H26, up 272% year-on-year from 5,000 in 1H25, with Chinese players accounting for 97%. About 65% of China’s shipments still went to entertainment, education/R&D and data collection, but industrial and commercial use cases are broadening. Half-size bipeds represented roughly 50% of volume and wheeled robots about 29%, particularly in commercial and industrial settings. The report expects more 1H26 pilots to convert to broader deployments from 2H26 and forecasts China humanoid shipments of 50,000 in 2026E and 446,000 in 2030E. Examples of operational progress include Ant Group’s Lingbot robots picking medicines in Shanghai pharmacies at night, T800 deployment for material handling in Luxshare’s Suzhou factory, and planned industrial rollouts by Hyundai and Schaeffler. The report identifies the robot “brain” and training data as the central global battleground. Companies are combining vision-language-action models, world models and reinforcement learning to reduce the marginal cost of teaching a robot a new task. Dyna’s DYNA-2 was trained primarily on more than one million hours of egocentric human video and reports a scaling relationship between human-video training and robotic capability. Skild AI’s S1 is designed to learn unseen tasks from videos or examples without task-specific post-training, while Generalist’s GEN-1.5 seeks to acquire simple manipulation skills from very short demonstrations without per-task fine-tuning. Morgan Stanley considers data the bottleneck, but highlights rising importance of data yield, diversity and deployment feedback; several players operate at around thousand-card compute scale and target one million usable data hours by end-2026 or 2027. Figure’s crowd-sourced Figure Index had 260,000 app downloads, more than 44,000 weekly active users and 16 million videos, and Figure plans to spend more than US$1 billion on data and compute over the next 12 months. Capital formation and public-market benchmarks are developing alongside the technology. Unitree’s Shanghai IPO raised Rmb6 billion; after a first-day gain of 460% and a subsequent 30% decline, its market capitalization was Rmb237 billion (US$35 billion), implying 80x and 48x price-to-sales on FactSet’s 2026 and 2027 revenue estimates of Rmb2.9 billion and Rmb4.9 billion. Morgan Stanley views the listing as a valuation benchmark for other Chinese humanoid companies. XPENG Robotics raised more than US$900 million at a valuation above US$6.3 billion, which the report describes as giving the business an external investor base and a more defensible carve-out valuation reference. Financing for robot-intelligence companies also supports the report’s view that capital is increasingly targeting hardware-agnostic intelligence rather than only robot bodies. Policy support is another commercialization accelerator, particularly in China. Morgan Stanley notes that China’s 15th Five-Year Plan made robotics a strategic emerging industry, while a June 2026 initiative targeted 10,000-level deployment capacity and more than 100 high-value applications by end-2026. National and local policies also call for operational training sites, funds, standards and application subsidies. The report expects sustained national and local support as commercialization begins, helping early adoption and reinforcing China’s development flywheel. It also notes policy activity in South Korea, Japan, the United States and the United Kingdom, while flagging that physical AI raises data-security, surveillance and dual-use national-security concerns. Morgan Stanley’s long-term adoption model projects cumulative global humanoid adoptions of roughly 28.1 million by 2036, 138.5 million by 2040, 430.4 million by 2044 and about 1.019 billion by 2050. Of the 2050 total, it estimates 935.1 million commercial and 84.2 million household adoptions; China accounts for about 302.3 million and the United States for about 77.7 million. Based on a six-year replacement cycle and falling average selling prices, the report estimates annual global humanoid revenue of US$336 billion by 2035, US$2.0 trillion by 2040 and US$7.5 trillion by 2050. High-income-market ASPs are assumed to decline from US$200,000 in 2024 to about US$75,000 by 2050, while lower-cost markets decline from roughly US$50,000 to US$21,000. For market tracking, the equal-weighted Humanoid 100 was up 47% since its February 6, 2025 inception as of August 28, 2026, outperforming the S&P 500, MSCI Europe and MSCI China but underperforming MSCI Korea and MSCI Taiwan. In contrast, the equal-weighted China Humanoid Value Chain was down 1.3% month-to-date and down 21.8% year-to-date, versus MSCI China’s -0.5% and -9.1%, respectively. Morgan Stanley organizes the investment universe into Brain, Body and Integrator categories, but the report’s central message remains that commercialization and productivity verification—not demonstrations or valuation alone—will determine the next stage of progress.

Analysis framework

Morgan Stanley combines event observations from the World Humanoid Robot Games with shipment data, company and financing developments, model and data updates, policy tracking, public-equity performance screens and a long-term adoption and ASP model. It evaluates progress by asking whether robots can complete variable, multi-step tasks autonomously and whether pilots can convert into scalable deployment.

Methodology notes

• Industry AnalysisSupply-demand framework

  Humanoid adoption and market-sizing model

  The report estimates adoption by geography, income group and commercial versus household use, then combines unit demand, replacement demand and selling-price assumptions to estimate revenue.

• Industry AnalysisUpstream-Midstream-Downstream Transmission

  Brain, Body and Integrator value-chain mapping

  Morgan Stanley groups public companies by semiconductors/software, industrial components and full-robot developers to show where humanoid demand and technology progress may flow through the value chain.

• Valuation methodsPS valuation

  Price-to-sales comparison

  The report uses Unitree’s implied 2026 and 2027 price-to-sales multiples as a valuation reference point for the emerging Chinese humanoid sector.

Asset mapping & comparison

Structured mapping from thesis to named assets (strengths, weaknesses, peers, risks).

• Humanoid 100

  Morgan Stanley’s global screen of public companies materially exposed to humanoid robotics.

  Strengths

  Broad exposure across Brain, Body and Integrator categories; up 47% since inception as of August 28, 2026.

  Weaknesses

  Underperformed MSCI Korea and MSCI Taiwan since inception.

  Comparison

  Outperformed the S&P 500, MSCI Europe and MSCI China.

  Risks

  Performance reflects price returns only and does not include dividend reinvestment or transaction costs.

• China Humanoid Value Chain

  Morgan Stanley’s equal-weighted list of 44 China-related humanoid companies and private players.

  Strengths

  China has leading shipment share and significant policy support for deployment, applications and training data.

  Weaknesses

  The value chain was down 21.8% year-to-date as of August 28, 2026.

  Comparison

  Underperformed MSCI China year-to-date by 12.7 percentage points.

  Risks

  Commercial use remains early and many shipments are still concentrated in entertainment, education/R&D and data collection.

Key data

• Global humanoid shipments19,000 units in 1H26Up 272% year-on-year from 5,000 units in 1H25; Chinese players accounted for 97%.
• China shipment forecast50,000 units in 2026E; 446,000 units in 2030EMorgan Stanley expects pilots to broaden into deployments from 2H26.
• Global cumulative adoption forecast28.1m in 2036; 138.5m in 2040; 430.4m in 2044; 1.019bn in 2050Morgan Stanley estimates; 2050 total includes 935.1m commercial and 84.2m household adoptions.
• Global annual humanoid revenue forecastUS$336bn by 2035; US$2.0tn by 2040; US$7.5tn by 2050Based on a six-year replacement cycle and declining ASP assumptions.
• Humanoid 100 performance+47% since February 6, 2025Equal-weighted, adjusting for additions and deletions, as of August 28, 2026.
• China Humanoid Value Chain performance-1.3% MTD; -21.8% YTDEqual-weighted as of August 28, 2026, versus MSCI China at -0.5% MTD and -9.1% YTD.

Impact & implications

The report argues that the sector is entering a more consequential phase: technical demonstrations must become autonomous, adaptable and productive deployments. Data availability, model transferability, deployment feedback, policy support and pilot conversion are presented as the main forces shaping commercialization and the relative positioning of Brain, Body and Integrator participants.

Risks

• Training data remains the sector’s key bottleneck, and models must prove they can transfer reliably to real robot tasks.
• Most deployment sites still operate only tens of robots, so pilot conversion into scalable, productive use remains unproven.
• Physical-AI systems raise data-security, surveillance and dual-use national-security concerns.
• Some reported financing, acquisition and deployment developments remain subject to incomplete details or ongoing discussions.

What to watch

• Whether 1H26 pilots convert into broader deployments from 2H26.
• Evidence that robots can complete real-world multi-step tasks autonomously and at useful productivity.
• Progress toward one million usable data hours, data quality improvements and deployment-feedback loops.
• Commercial evaluations and rollouts of foundation models such as DYNA-2, Skild S1 and GEN-1.5.
• Execution of China’s end-2026 targets for 10,000-level deployment capacity and more than 100 high-value applications.