Written by | Wu Xianzhi Edited by | Wang Pan
As Unitree Robotics released its Q3 2026 financial report on October 29, the embodied intelligence industry was navigating a subtle cyclical period.
Recent reports indicated that after entry points for humanoid robots into the secondary market tightened, this pressure further transmitted to the primary market. Some investors holding legacy shares of a certain embodied AI company valued at CNY 10 billion were seeking buyers at half price. When the waves at the crest are no longer turbulent, areas previously shallow become easily exposed to sunlight.
In September, Mech-Mind Robotics founder Shao Tianlan publicly questioned on WeChat Moments whether some embodied AI companies were leveraging data collection centers to manufacture unsustainable revenue, thereby supporting valuations and IPO expectations. Interestingly, while disclosing its Q3 report, Unitree also revealed a related-party transaction involving a data collection center.
From information disclosed in the prospectus and Q3 financial report, combined with Unitree’s and the industry’s movements this year, it is easy for outsiders to see that this leading embodied AI company is undergoing a transformation.
In the prospectus, Unitree’s robots relied on hardware plus "little brain" (low-level control) modules for profitability. After listing, Wang Xingxing repeatedly expressed the view that Unitree needs to address its shortcomings in model development.
On October 22, during his speech at the 2026 World Physical AI Conference and Unitree Partner Conference, Wang Xingxing’s theme was "Believe, believe in AI as much as possible." In the speech script of over 1,800 words, "AI" appeared 25 times, while "embodied" and "robot" together appeared only 18 times.
Considering the industry narrative, timing, and company situation, the significance of Unitree’s quarterly report goes without saying.
Humanoids Can Generate Profit, But Margins May Not Be High
The data in Unitree’s financial report answers two questions of great concern to the industry: Do not doubt that humanoid robots can make money, but do not harbor unrealistic fantasies about profits.
Since last year, Unitree has launched multiple robots, including the small-to-medium humanoid robot R1 and the full-size humanoid robot H2. On October 21, they also unveiled the "Kung Fu Legend" H2-X humanoid robot. However, shipment data for these models is not yet available. The only mass-produced product verifiable is the small-size humanoid robot G1, which Wang Xingxing previously stated had a cumulative production volume of 10,000 to 20,000 units.
As China’s "first humanoid robot stock," the G1 is the model capable of influencing financial data. Since Q3 reports do not mandate detailed product disclosures, it is difficult to quantify further.
However, by combining the shift from quadrupeds to humanoids in 2026 with revenue and profit figures, we can discern broader trends. As of September 30, Unitree’s single-quarter revenue reached CNY 665 million, a year-on-year increase of nearly 70%. This revenue growth was likely achieved through volume-driven pricing strategies. Two pieces of evidence support this: first, the net profit attributable to shareholders excluding non-recurring items was CNY 105 million, a year-on-year decline of 17.4%.
Second, deducting the first-half data yields the third-quarter figures. We note that the estimated gross margin for core business operations was approximately 50.6%, lower than the 56% in the first half of this year, and lower than margins in 2024 and 2025, exceeding only the 44.2% recorded in 2023.
The prospectus had already provided directional guidance earlier, noting that the starting price of the flagship humanoid product G1 dropped from CNY 99,000 to CNY 85,000, while the price of the "small airplane" R1 Air robot (a consumer-grade entry-level lightweight bipedal humanoid targeting C-end users and research) fell to CNY29,900.
This presents a somewhat counter-intuitive signal: while the proportion of high-margin humanoid robots continues to rise, the per-unit profit of humanoid robots has been continuously declining in recent years. The reasons are multifaceted, with intense competition being an unavoidable factor.
For mid-range general-purpose bipedal humanoids like the Unitree G1, targeting developers and research scenarios, competitors include AgiBot A2, Fourier GR-2, and UBTECH Walker S1. Prices for several of these models are quite close, differing by only thousands of yuan. For family-oriented consumer-grade entry-level lightweight products, Unitree G1 (or Go2) faces competition from models such as AgiBot A2 Youth Edition and Fourier GR-1 Lite.
The proliferation of similar models and continuous price drops have more or less affected Unitree’s gross margin levels. After national subsidies, the price of Bumblebee (a specific model reference implied by context or typo in source, translated as is if unclear, but here 'Bumi' likely refers to a competitor or internal code, assuming generic context or keeping phonetic if unknown - *Correction*: Context suggests 'Bumi' might be a typo for a competitor or specific unit. Given constraints, I will translate '布米' as 'Bumi' or assume it refers to a low-cost unit. Let's use 'Bumi' as proper noun). The post-subsidy price of Bumi has dropped to CNY 9,418, essentially hitting the floor for this niche category. Hitting the price floor also opened up the home market. An industry insider believes that data collection in home scenarios must mature before entering homes; therefore, current consumer-grade humanoid products cannot yet support the commercialization of the "brain" (high-level AI).
Amid increasingly diverse body products and intensifying price competition, Unitree has bet resources on the model side. The prospectus explicitly states that nearly half of the CNY 4.2 billion raised will be invested in intelligent robot model R&D.
In Q3, the company’s R&D investment exceeded CNY 127 million, surging nearly 250% year-on-year, approaching the sum of the first two quarters of this year. The company stated that it is continuously increasing R&D investment in embodied intelligence, particularly for embodied large models and computing power.
Reviewing Unitree’s actions throughout the year reveals that the company is gradually filling its gaps in model capabilities.
In January, Unitree open-sourced the VLA model UnifoLM-VLA-0, enabling robots to understand tasks in front of them and perform physical operations, effectively adding the "hands" to the model side.
In September, they open-sourced the new-generation general-purpose humanoid foundation model UnifoLM-WLA-1.0. This model primarily addresses generalization and whole-body coordination, completing 64 tasks including tabletop and full-body operations with a single model, supporting grippers and two types of dexterous hands. This represents a phased achievement in Unitree’s effort to supplement its "brain."
On October 22, Wang Xingxing revealed that the company is pre-researching the Physical AI Robot Self-Evolution V1.0 Large Model, aiming to integrate AI from the underlying R&D stage into robot manufacturing.
In Unitree’s financial report, we noted that after focusing on humanoid robots, the company’s cash collection quality has steadily improved. The ratio of cash received to cash paid for goods/services increased from 107.6% in 2025 to 111.3%. Although there is still a distance between consumption/research scenarios and productivity, the high quality of cash collection indicates that humanoid robotics is not a bad business.
Data Collection Is a Necessary Expenditure
Changes in Unitree’s inventory reflect the company’s transition from concept to commercial mass production.
In Q3, Unitree’s inventory value stood at CNY1.02 billion, an increase of 177.2% from the beginning of the year and 51.9% from the end of the first half.
Analyzing Unitree’s financial data, we can roughly discern a typical capital structure for hardware product stocking. During the same period, accounts payable rose from CNY 245 million to CNY 693 million, a net increase of CNY 448 million, covering 69% of the inventory value.
This set of data is familiar to automakers: upstream suppliers need to advance part of the component costs, and the remaining gap can only be covered by other working capital and the company’s own cash generation. This quarter, Unitree had CNY 194 million in operating cash inflow, covering about 30% of the CNY 652 million increase in inventory.
Funding may not be Unitree’s most pressing concern currently. For the company and the entire industry, the primary issue remains solving effective data supply.
In his October 22 speech, Wang Xingxing mentioned that top-tier large model companies have largely achieved self-evolving models internally. "The models collect data themselves, write code themselves, and humans only provide guidance. This logic can be fully transferred to robots."
This indicates that Unitree hopes to be deeply involved from the initial data collection stage. Coincidentally, beyond the financial report disclosure, Unitree also announced a related-party transaction involving data collection.
The announcement showed that Hangzhou Embodied Intelligence Application Pilot Base Co., Ltd. conducted a capital increase and share expansion. Unitree Technology Co., Ltd. planned to subscribe CNY 20 million using funds from its wholly-owned subsidiary, Hangzhou Unitree Robot Co., Ltd., to maintain its 10% equity stake.
Public information shows that the establishment of the Hangzhou Pilot Center aims to address the current data supply shortage in the embodied intelligence industry. The center consists of a capability supply center, model training center, scenario verification center, and testing certification center, forming a complete system of computing power, data, models, and scenario applications.
The first batch of resident enterprises includes 18 companies, comprising embodied AI firms like Xinghaitu, GALBOT, and Moushen Intelligent, as well as mature tech giants like Huawei, Alibaba, and Moore Threads.
Li Xingteng, Deputy General Manager of the base, previously gave an insightful metric for data collection in an interview with authoritative media: one humanoid robot requires 1–2 data collectors to cooperate for full-body motion data collection. A simple action requires "feeding" tens of thousands of real motion data points, and the effective data rate is less than 50%. This is already considered a high standard compared to teleoperation-based collection during the same period.
The center has produced tangible results. It is understood that based on data collected from Unitree robots, the first batch of real-machine datasets has been open-sourced.
Another piece of data in the announcement reflects the importance attached by enterprises and the government to the data production phase.
In Kunyuan Asset’s valuation case, it was explicitly stated that audited owner’s equity as of December 31, 2025, was CNY 47.6581 million. By September 30, 2026, unaudited data showed owner’s equity remaining at CNY 32.4534 million, representing a value change of over CNY 15.2 million.
When market capital has not yet entered, one cannot expect this public infrastructure-style data collection center to directly generate economic benefits immediately; it requires industrial and even higher-dimensional resource injection. The Hangzhou Pilot Center is one such outcome. Unitree stated that the current capital increase for the pilot base meets the funding needs for construction projects and business development.
From a more practical perspective, before the market is fully opened, the company already possesses a high-standard data production workshop to support model iteration.
Unitree Endorses the Industry
When Unitree’s prospectus was disclosed, people held great expectations for this domestic "first humanoid robot stock."
The day of the listing coincided with the first day of WRC 2026. All industry insiders we contacted were extremely concerned about its performance in the secondary market, almost all hoping the stock price would climb steadily. One practitioner even believed that Unitree equals the industry ceiling.
After two months of consolidation, Unitree’s transformation has gained more informational support, and the market, returning to rationality, has begun re-evaluation.
In 2025, scientific research and data collection were the main scenarios absorbing humanoid robots. That year, Unitree’s humanoid robot sales exceeded quadruped robot sales for the first time, indicating that the product focus had completed its switch. Scientific research scenarios provide essential development-side value such as algorithm verification, data collection, and educational demonstrations for emerging body manufacturers.
However, this scenario indeed has a commercial ceiling, determined by project budgets. Since the beginning of this year, embodied body manufacturers have focused on industrial and commercial service scenarios because they know they need to persuade customers with ROI.
In the first half of 2026, global humanoid robot shipments grew nearly 300% year-on-year. In the first half, China shipped over 40,000 humanoid robots, accounting for 97% of the global total. Market data and Unitree’s operational data jointly constitute the current "Q3 report" of the humanoid robot sector.
If scientific research scenarios are viewed as a sub-item in the embodied intelligence series, then Unitree in 2025 was the absolute champion. Next, industrial, commercial service, and home scenarios are the finals, while every workshop and every sub-scenario serves as the preliminaries and semi-finals.