Goldman Sachs China humanoid robot research: The industry is shifting from "general imagination" to "specialized implementation", and 2026 may see "volume validation + expectation reset".

Goldman Sachs’ latest research indicates that China’s humanoid robotics industry is undergoing a strategic transformation from “general imagination” to “dedicated deployment.” This pragmatic approach, combined with significant advances in motion control capabilities and fast iteration cycles, is driving major manufacturers to set shipment targets for 2026-2027 at several times the scale of 2025.

According to Wind Chaser Trading Desk, Goldman Sachs analyst Jacqueline Du pointed out in her latest report that during a January 15-20 survey of eight humanoid robotics companies and related supply chain firms—including Unitree Robotics, UBTECH, Fourier, and CloudMinds—the team observed the industry shifting its focus from pursuing general capabilities to vertical scenarios such as security patrols, guidance services in public venues, and factory logistics sorting, all leveraging existing strengths in task planning, mobility, and interaction.

Based on feedback from major manufacturers and supply chain firms, Goldman Sachs estimates that global shipments of humanoid robots will reach approximately 15,000 to 20,000 units in 2025, with Chinese companies contributing the bulk of these shipments. Looking ahead to 2026-2027, leading manufacturers expect several-fold growth, scaling from hundreds to thousands of units in 2025 to several thousand or even tens of thousands.

Goldman Sachs notes that 2026 could be a pivotal year for “volume validation + expectations reset,” as investors watch to see whether milestones such as ‘a million robots’ are revised and how individual supply chain companies’ market shares and per-unit values evolve.

Aggressive shipment targets; capacity and testing present challenges

Goldman Sachs’ research shows global shipments of humanoid robots in 2025 to be between 15,000 and 20,000 units—close to Goldman’s earlier estimate of 20,000, and in line with third-party data predictions of 13,000-16,000 units. Chinese companies currently supply the vast majority, with demand mainly from research institutes, AI training for robots, education, entertainment performances, and data factories.

In this still-early-stage market, leading manufacturers have set ambitious growth goals for 2026-2027. Building on expected 2025 shipment volumes ranging from the hundreds to thousands, companies are setting targets of several thousand to tens of thousands for 2026-2027, signaling exponential growth.

Supporting factors for these growth expectations include the maturing of the supply chain, optimized cost curves, and expansion of application scenarios. However, Goldman Sachs highlights that realizing these targets will face challenges from ensuring production consistency and the multi-staged testing processes inherent in this emerging industry.

Significant improvements in motion control; iteration cycles shortened to 6-8 months

During live product demonstrations, Goldman Sachs analysts observed substantive progress in humanoid robots’ motion control. Both wheeled upper body platforms and full bipedal systems displayed stronger robustness and flexibility, showing marked improvements compared to the previous year.

One manufacturer claimed to have achieved “cerebellum-level” whole-body control, providing two practical benchmarks: robots can navigate in unmapped terrain, and can be controlled remotely as a whole rather than in segmented upper and lower body operations.

Supply chain integration capabilities are accelerating product iteration. Multiple companies revealed that iteration cycles for humanoid robotic platforms have been reduced to about 6-8 months per generation. This rapid pace is largely attributed to 80%-90% of components being self-designed, which is vital for seamless hardware-software integration and optimizing performance within compressed R&D and testing timeframes.

Application focus shifts to 'dedicated deployment,' bypassing dexterity challenges

The “sim-to-real” gap remains a bottleneck for the industry. Currently, robot pre-training depends heavily on simulated and synthetic data, with an 80%-90% accuracy rate in simulation environments often dropping below 50% in real-world scenarios. Since gathering large-scale, high-quality real-world data and developing effective world models takes time, China’s leading humanoid robot developers are prioritizing “dedicated” commercial deployments.

These use cases include security patrols and guidance services in public places such as hotels, banks, museums, exhibition centers, car dealerships, and supermarkets—effectively harnessing existing capabilities in task planning, movement, and interaction, while avoiding the complexity of highly dexterous operations.

For industrial applications, robots that require dexterous hands or grippers are currently limited to tasks such as box moving and simple logistics sorting. This is mainly because AI struggles with unpredictable edge cases on factory floors. According to UBTECH, in sorting and logistics, once a robot reaches roughly 50% of a human worker’s productivity, clients are willing to invest, achieving payback in about two years (assuming ten-hour daily operation). Even in environments with acute labor shortages, a three-year payback is considered acceptable.

Data strategy becomes a key competitive edge; world models garner attention

Recently, humanoid robot makers are increasingly embracing standardized approaches, integrating with mature large language models (LLMs) and vision-language model (VLM) technology stacks from Alibaba (Qwen), Doubao, Tencent, and others. This strategy makes proprietary data engines a central point of differentiation in developing deployable robotic intelligence.

High-quality real-world data is seen as the main limiting factor for bridging developed hardware with scalable, practical applications. Thus, a “data recipe” arms race is underway, with differentiation driven by targeted end-applications.

Though all companies pursue data collection strategies, they use varied combinations of three primary data sources: human/expert remote teleoperation (high control but usually costly); simulated data (low incremental cost per sample but less realistic); and real-world video datasets (highest availability but may not translate to real-world accuracy).

Goldman Sachs found growing references to world model methodologies in this survey, which may endow robots with some common sense about their environments, shifting them from reactive agents to proactive intelligent systems capable of complex planning and adaptation.

Business models diverge: 2C focuses on experience, 2B on ROI

Different target markets have given rise to diverse profit models, mainly split into 2C (consumer-facing) and 2B (enterprise-facing) applications.

2C-oriented companies mainly focus on differentiated features and enhancing user experience, often emphasizing “emotional value,” capturing niche verticals, and commanding premiums via unique functions or interactions. Their goal is to create products that stand out in capabilities and user engagement.

In contrast, 2B-oriented firms anchor pricing to clients’ return on investment (ROI), usually by demonstrating how robots can boost productivity, improve efficiency, or reduce labor costs. UBTECH notes that in sorting and logistics, when robots reach roughly 50% of human worker productivity, clients are willing to invest, recouping their cost in about two years. Even in labor-starved environments, a three-year payout is acceptable, reaffirming the value proposition of automation in addressing critical operational challenges.

~~~~~~~~~~~~~~~~~~~~~~~~

The above highlights are from Wind Chaser Trading Desk.

For deeper insights, including real-time analysis and frontline research, please join [Wind Chaser Trading Desk · Annual Membership]

Risk Disclaimer and Legal StatementsThe market involves risk, and investments require caution. This article does not constitute personal investment advice, nor does it take into account individual users’ specific investment objectives, financial situations, or needs. Users should consider whether any opinions, viewpoints, or conclusions in this article are appropriate to their particular circumstances. Invest accordingly at your own risk.