Within just two years, Unitree Robotics has catapulted its annual humanoid robot sales from a mere 5 units to over 5,200, capturing roughly 30% of global shipments and establishing itself as one of the few profitable robot makers in this emerging sector.

On August 28, UBS released its inaugural coverage report detailing the company's prospects. The report deconstructs humanoid robot capabilities into three distinct layers: the "brain" (perception, reasoning, and task planning), the "cerebellum" (motion control and dynamic balance), and the "body" (actuator performance and reliability). The conclusion is clear: Unitree holds a significant edge in the "cerebellum plus body" categories, yet the "brain" — the capacity for AI generalization — remains an industry-wide bottleneck that has yet to be solved.

UBS analysts noted that the current valuation of 54 times the projected 2027 price-to-sales ratio already reflects the company's status as a hardware leader and anticipated 53% revenue compound annual growth rate from 2026 to 2028. For the stock to climb further, more tangible progress on the "brain" front and a sustainable path to profitability conversion are needed.

The company debuted on the STAR Market on August 19, marking the first A-share listed humanoid robot maker. It raised 6.099 billion yuan in actual funding, with its share price hovering around four times the IPO price of 150.8 yuan as of the report date. From 2022 to 2025, revenue grew at a 140% compound annual rate, while gross margin climbed from 43.6% to 60.3%, culminating in a net profit of 278 million yuan in 2025.

In 2025, industrial applications of humanoid robots constituted just 3% of global demand, and a whopping 76% of Unitree's humanoid revenue came from scientific research and education. The company has directed roughly 39% of its total IPO proceeds toward embodied AI development and forged a strategic partnership with DeepSeek in August.

Hardware Dominance — A Full-Stack Moat in Actuator Self-Development and Motion Control

Current buyers of humanoid robots are less concerned about whether these machines can "think clearly" and more focused on whether they can "move effectively." Commercial performances demand backflips and synchronized group dances, research labs need high-performance hardware platforms, and data collection centers require stable execution carriers — all scenarios where demands on motor skills and hardware reliability far outweigh AI generalization capability.

This is precisely where Unitree's defensive moat is built. The company independently develops actuators, motor drives, control systems, structural components, and LiDAR sensors. Actuators account for over 40% of the BOM cost in a humanoid robot. While most robot makers source complete joint modules from third-party suppliers, Unitree designs its own components, procures parts externally, and assembles them in-house. Its self-developed M107 motor delivers a peak torque of 360 N·m and a maximum pulling force of 10,000 N. The technological path relies on a QDD (quasi-direct drive) architecture — high-torque-density motors combined with low-ratio planetary gearboxes and high-bandwidth force control — striking a balance between dynamic response and engineering feasibility, making it well-suited for high-dynamic maneuvers like running and jumping.

On the motion control front, the company employs deep reinforcement learning and large-scale parallel simulation training to close the loop between multimodal perception and motor control. The H1 standard version reaches a maximum speed of 3.3 meters per second, with a theoretical capability exceeding 5 meters per second, while a competition-tuned variant peaks at 10 meters per second. A prototype dubbed "Superman," unveiled in August 2026, claims a top speed of 12.66 meters per second — surpassing the human running world record — though it has yet to be independently verified by third parties. The company has also wowed audiences with two consecutive Spring Festival Gala robot dance performances and topped the medal table at the inaugural World Robot Games with 11 medals, underscoring the battle-tested maturity of its whole-body coordination and multi-robot formation control.

This capability has translated directly into shipment volumes. Global humanoid robot shipments reached approximately 18,315 units in 2025, with Unitree leading the pack at 5,500 units (5,215 recognized as revenue). Cumulative quadruped robot sales exceed 33,000 units, including 23,037 units in 2025 alone, capturing about 28% of the global share. Overseas revenue contributes over 40% of total sales. The momentum continued into the first half of 2026: humanoid shipments hit roughly 5,900 units, maintaining a global share around 31%, while revenue surged 48.5% year-on-year to 1.15 billion yuan. During the same period, global humanoid robot demand reached 19,100 units, a 272% year-on-year jump.

ASP Halved, Profits Rose — The Real Moat Lies in Cost Structure

In 2024 and 2025, Unitree's average humanoid robot price dropped by 56% and 36%, respectively, while quadruped prices fell by 16% and 6% over the same periods. Yet gross margin climbed from 43.6% in 2023 to 60.3% in 2025. In hardware, it's uncommon to see margins expand while prices fall alongside volume growth.



The table below illustrates the financial trajectory:

Metric 2023 2024 2025 2026E 2027E 2028E

Revenue (100M yuan) 1.59 3.93 16.99 27.42 46.22 63.98

Net Profit (100M yuan) -0.11 0.95 2.78 6.04 8.92 12.21

Gross Margin 44% 57% 60% 58% 58% 58%

This outcome is driven by lower BOM costs achieved through full-stack in-house development — vertical integration from motors to structural parts removes the profit cut typically taken by external suppliers. The industry's largest shipment scale also affords significant procurement bargaining power. Additionally, the product mix has been optimized: humanoid robot revenue share climbed from 27% in 2024 to 51% in 2025, and since humanoid gross margin (63%) exceeds that of quadrupeds (57%), this shift has directly boosted the overall margin profile.

Peer comparison makes the point even clearer. In 2025, gross margins stood at 60% for Unitree, 38% for UBTech, and 46% for JAKA. On the expense side, Unitree's operating expense ratio was 31.8% (including about 350 million yuan in equity incentives); stripping that out, its management expense ratio is just 2.97%, far below the 47%-75% range typical of its peers. During the industry's early phase, Unitree focused R&D resources on hardware and motion control rather than foundation models, resulting in a lower R&D expense ratio than UBTech and JAKA, but its hardware competitiveness has translated into more orders and stronger profitability.

However, Unitree's profit margins faced pressure in the first half of 2026. Net margin dipped to 24%, partly due to a factory relocation that dragged gross margin, combined with one-time Spring Festival brand promotion costs and a 150% year-on-year spike in R&D expenses — the latter being the financial reflection of a strategic pivot toward the "brain." UBS estimates that increased R&D will weigh on net margins in 2026-2027, but margins are expected to stabilize above 19% from 2028 to 2030, with net profit growing at roughly 57% compound annual rate from 2025 to 2030.

Bridging the "Brain" Gap — DeepSeek Alliance and 2 Billion Yuan R&D Investment

Industrial applications account for just 3% of global humanoid robot demand for a straightforward reason: insufficient AI generalization means robots still struggle with unstructured tasks. The industry consensus is that humanoid robots are currently in a phase akin to large language models before ChatGPT — embodied intelligence models continue to evolve, but the breakthrough application inflection point remains some time away.

Unitree is pursuing a dual-track approach: developing both world models (WMA) and vision-language-action models (VLA), having open-sourced UnifoLM-WMA-0 and UnifoLM-VLA-0 in September 2025 and January 2026, respectively. About 56% of IPO proceeds are channeled into R&D, with roughly 39% going directly to embodied AI models and core algorithms. The company projects R&D investment exceeding 2 billion yuan over the next two to three years, above the industry average. With an estimated net cash position of approximately 8 billion yuan in 2026, the company is financially equipped to sustain this spending cycle.

In August 2026, DeepSeek entered into strategic and capital partnerships with Unitree — securing 933,400 shares in the IPO's strategic placement for approximately 141 million yuan, subject to a 36-month lock-up period, alongside a signed Strategic Cooperation Memorandum covering general AI, high-performance robotics, and AI foundation models. The 36-month lock-up signals a multi-year industrial collaboration rather than a short-term financial play. DeepSeek's large model capabilities combined with Unitree's robotics hardware, motion control, and physical-world data could accelerate the transition from "can move" to "can understand, decide, and execute." That said, early-stage R&D and compute costs may rise, and integration complexity with commercialization timelines still carries uncertainty.

On the capacity front, a new production base will add 75,000 humanoid and 115,000 quadruped robot units annually once fully operational. The company is also building an ecosystem loop spanning "hardware, toolchains, developers, applications, and data," with the core logic of attracting developers through an installed base, enriching applications and data through developer contributions, and continuously iterating products and lowering costs based on real-world data. Across 23 Chinese provinces, 64 humanoid robot data collection centers are operational, with 26 more under construction, consistently feeding AI model training. In an optimistic scenario, global humanoid robot demand is projected to grow from 18,300 units in 2025 to 325,000 units by 2030, a compound growth rate exceeding 70%, with a market size potentially reaching $16 billion.

For now, Unitree has proven it can build the best "body" in the business.

The only question that remains: can this body be equipped with a truly intelligent "brain."