China’s Humanoid Robot Reality Check: Unitree’s Valuation Reset and AgiBot’s IPO Gamble
Unitree Robotics shares have lost more than half their value since listing, while AgiBot prepares a Hong Kong IPO at a price-to-sales multiple that rivals Unitree's own debut valuation—exposing a fault line between two radically different bets on how China's embodied AI industry will monetize.
The divergence crystallized on Sept. 10, 2026, when Unitree Robotics closed at RMB 498.55 per share, slipping below the psychologically critical RMB 500 threshold. Market capitalization stood at RMB 201.6 billion (US$28 billion), down from an opening-day peak of RMB 444.9 billion (US$61.8 billion) when the stock debuted on Shanghai's STAR Market. The same week, rival AgiBot confirmed its intention to list on Hong Kong's Chapter 18C exchange at a target cornerstone valuation of HK$40–50 billion (US$5.1–6.4 billion)—positioning itself, despite being unprofitable, at a price-to-sales ratio of 32.8x–40.9x based on its 2025 revenue of RMB 1.05 billion (US$145.8 million). That multiple nearly mirrors the 36x P/S at which Unitree itself went public. Capital, in other words, has not yet discounted the sector's story—but secondary-market investors clearly have.
Unitree's Revenue Mix Reveals Why the "Expensive Toy" Label Refuses to Die
The bear case against Unitree is not about technology. On Sept. 7, 2026, the company released its UnifoLM-X2-1.0 world-action foundation model, enabling its H1 humanoid to execute unprompted punching, blocking, and dynamic evasion in a live demonstration—a genuine milestone in autonomous physical reasoning. Two trading sessions later, the stock fell further.
The market's skepticism is anchored in Unitree's 2026 interim financials. Revenue growth decelerated to 48.54% year-on-year in the first half, a sharp step-down from the 335% full-year expansion recorded in 2025. More damaging: non-GAAP net profit fell 19.34% in the same period, and first-quarter 2026 non-GAAP net profit dropped 52.55% year-on-year. The company's own updated IPO risk disclosures—revised between the March 2026 filing draft and the May 2026 listing committee submission—elevated "year-on-year net profit decline" to the top of all special risk clauses, a rare act of public self-flagellation.
The structural problem is visible in the revenue breakdown. In 2025, humanoid robots surpassed quadruped robots to account for more than 51% of Unitree's total revenue of RMB 1.699 billion (US$236 million). Yet within that humanoid segment, over 70% of sales flowed to universities and research institutes, with genuine industrial deployment representing less than 3% of revenue. The company's most prominent brand moments—16 H1 units performing a folk dance choreographed by Zhang Yimou at China's 2025 Spring Festival Gala, following an earlier appearance by 24 quadruped robots in 2021—are spectacular demonstrations of locomotion capability. They are not proof of a functioning commercial payback loop.
This creates a compounding constraint unique to Unitree's STAR Market listing. The exchange imposes hard metrics around earnings consistency, operating cash flow, and profit stability. Those rules force Unitree to prioritize near-term margins over long-cycle R&D bets—precisely the opposite of what closing a 70%-to-3% research-to-industrial revenue gap requires. Research and development spending rose from RMB 49.95 million (US$6.9 million) in 2023 to RMB 145 million (US$20.1 million) in 2025, but the pace remains modest relative to the scale of the AI capability gap the company now admits it must close. Unitree's IPO prospectus allocates RMB 2 billion (US$277.8 million) to model development versus RMB 1.1 billion (US$152.8 million) to hardware—a strategic pivot in writing that has yet to show up in the income statement.
AgiBot's IPO Faces a Cooling Sector Temperature Even as Its Capital Architecture Offers Partial Insulation
AgiBot Robot was founded in February 2023 by Deng Taihua, a former Huawei vice president who oversaw the Kunpeng and Ascend AI computing ecosystems for two decades. Co-founder, president, and CTO Peng Zhihui — known online as "Zhihui Jun," a former Huawei "genius youth" hire who worked on Ascend AI chips—provides the company with organic media reach unusual for a three-year-old hardware startup. More than half of AgiBot's senior management team carries Huawei pedigree, and the operating playbook—internal horse-racing, ecosystem bundling, aggressive sales incentives—is recognizably derived from its founders' former employer.
That institutional credibility accelerated fundraising in a way Unitree's purely technical reputation could not replicate. AgiBot has closed 11 financing rounds to date, drawing commitments from more than 50 investors including Tencent, BYD, Sequoia China, Hillhouse Ventures, and CASSTAR. In July 2025, AgiBot paid RMB 2.1 billion (US$291.7 million) to acquire a 63.62% stake in listed materials company Swancor Advanced Materials via a combination of agreement transfer and tender offer, establishing a dual-market capital platform: the parent entity pursues the Hong Kong 18C listing while the A-share vehicle operates independently.
The more analytically significant structure is AgiBot's subsidiary spin-off strategy. The company has carved out five independent subsidiaries—covering robotics leasing (Botshare), data services (Mi Bee Body Intelligent), dexterous hands, cleaning robots, and quadruped robots—each raising capital independently. Qingtian Rent has reached a standalone valuation of RMB 7 billion (US$972.2 million), qualifying as a unicorn. In August 2026, Qingtian Rent fully acquired Aizooji, a legacy 3C consumer electronics rental platform, to build out the operational infrastructure for robot-as-a-service (RaaS) deployment in industrial parks and factory floors. The logic is deliberate: strip capital-intensive, low-margin leasing and data businesses into subsidiaries, keep the parent company's narrative focused on high-multiple "universal AI brain" positioning.
The commercial flywheel AgiBot describes is three-layered: hardware margin on robot unit sales; data revenue from real-world deployment telemetry fed back into model training and sold externally; and leasing commission income through Qingtian Rent, including depreciation buybacks. Underpinning all three layers is what the company calls its "partner-first" strategy—equity co-investment with customers and suppliers. AgiBot has reached capital-level partnerships with more than 40 listed companies and has established at least 17 joint ventures to penetrate specific deployment scenarios. Supply-chain partners including PIA Automation, Lens Technology, and ASD have formed joint ventures with AgiBot, embedding procurement relationships within the equity structure.
The critical question for post-IPO performance is whether this architecture constitutes genuine demand or simply demand manufactured through capital obligations. When an investor is also a buyer, the commercial signal embedded in an order book becomes ambiguous. This is the core analytical challenge facing any investor attempting to assess AgiBot's revenue quality ahead of listing.
Parsing the Shipment Numbers: How Much Is Real Market Demand?
The duopoly's competing shipment claims illustrate the opacity problem directly. According to data published by Omdia, AgiBot shipped more than 5,100 general-purpose humanoid robots in 2025, capturing approximately 39% of global market share—the highest of any single manufacturer. Unitree disputes that ranking, citing its own prospectus figures: more than 5,500 humanoid units shipped and more than 6,500 units completed production in 2025. When AgiBot announced it had crossed the 10,000-unit cumulative production milestone, Unitree simultaneously announced approximately 11,000 units produced. The numbers are close enough that neither company can claim decisive hardware leadership.
What neither figure clarifies is the proportion of shipments driven by end-market pull versus capital-relationship push. For Unitree, the 70%-plus concentration in universities and research institutes suggests a buyer base that is price-sensitive to government budget cycles and institutionally incapable of scaling deployment at industrial velocity. For AgiBot, the joint-venture and equity-partnership structure means a meaningful portion of its order book flows from entities that have financial incentives to purchase regardless of immediate operational return on investment.
TrendForce projects China's humanoid robot production volume will grow 94% year-on-year in 2026, with Unitree and AgiBot together accounting for nearly 80% of total domestic shipments. That concentration underscores how early the market remains: two companies controlling 80% of output in a market where genuine industrial deployment—robots performing repeatable, economically justified tasks outside of demonstration contexts—represents a fraction of total units in the field.
The industry's honest accounting problem is that "shipment" and "deployment" are not synonyms. A robot delivered to a trade show booth or leased to a property developer for lobby aesthetics generates revenue but not the kind of operational data that trains a general-purpose model toward commercial-grade reliability. Until the industry develops standardized disclosure around revenue from sustained industrial contracts versus one-time research and showcase sales, shipment figures will remain a proxy metric that obscures more than it reveals.
Converging "Big Brain/Small Brain" Routes May Accelerate—But Not Guarantee—a Commercialization Inflection
The most consequential development of September 2026 is not Unitree's stock price. It is the collapse of the clean boundary between the two companies' technical philosophies.
Unitree built its reputation on "small brain" excellence: hardware locomotion, joint control, and physical stability that allows its robots to perform backflips, sprint, and now engage in autonomous combat. AgiBot's founding thesis was "big brain" primacy: a general intelligence layer built on foundation models, scene understanding, and data flywheels that would eventually command any compliant hardware body. These were genuinely different bets on where the value in embodied AI would accumulate.
That distinction is eroding. Unitree's September 2026 release of the UnifoLM-X2-1.0 world-action model—and the prospectus capital allocation that puts AI model development ahead of hardware investment for the first time—signals that the company recognizes it cannot sustain premium valuation as a pure-hardware vendor. Simultaneously, AgiBot's aggressive investment in Qingtian Rent's operational infrastructure and the Aizooji acquisition signal that a "big brain" narrative alone cannot generate the industrial deployment density needed to train models at scale. Both companies are being forced toward the same convergent architecture: reliable hardware as the physical substrate, foundation model intelligence as the value-capture layer.
Whether convergence accelerates the commercialization inflection depends on which bottleneck is harder to solve. The evidence from Unitree's revenue mix suggests the limiting constraint is not locomotion quality—it is use-case definition and industrial integration. Factories do not buy robots because robots can do backflips. They buy robots when the total cost of robotic labor, including integration, maintenance, and retraining, falls below the cost of human labor for a specific, repeatable task. That calculation currently does not close for most humanoid applications. Unitree's CEO has acknowledged this implicitly by redirecting IPO proceeds toward model research; AgiBot's Deng Taihua has been explicit, framing the company as a "long-distance runner" that "does not rush toward positive net profit."
In the first half of 2026, total disclosed financing into China's embodied AI sector reached RMB 104.1 billion (US$14.5 billion)—nearly double the full-year 2025 figure. Structurally, more than half of that capital flowed to "big brain" companies focused on AI software and algorithms, while hardware-body manufacturers received less than 20%. The capital vote is clear: investors believe software and data will capture more value than hardware over the medium term. But software value in robotics is only realizable through hardware deployment at scale—a chicken-and-egg dynamic that neither company has yet resolved.
Capital Can Raise Valuations, But Cannot Skip the Commercialization Exam
The parallel trajectories of Unitree and AgiBot represent two legitimate but incomplete strategies for navigating an industry that has not yet found its iPhone moment. Unitree's STAR Market listing has imposed earnings discipline that limits its ability to absorb the losses required for transformative R&D; the stock's 50%-plus decline from its opening price is the market's verdict on a company that monetizes primarily through institutional showcase sales. AgiBot's capital-network model has created an order book that is difficult to disaggregate from its investment relationships, and its "358 Vision Plan"—RMB 10 billion (US$1.39 billion) in revenue by 2027, RMB 100 billion (US$13.9 billion) by 2030—will require sustained external funding at a time when sector sentiment in secondary markets is visibly cooling.
The sector's IPO wave—Unitree on STAR Market, AgiBot targeting Hong Kong 18C, DEEP Robotics with a STAR Market application accepted, and more than 20 additional robotics companies in active capital-market preparation—means the embodied AI industry will face its commercial credibility test in public, with quarterly disclosure requirements and impatient institutional shareholders.
The question is not which company wins the duopoly. It is whether either company—or the sector collectively—can demonstrate before the next funding cycle that robots are leaving laboratories and trade show stages and generating auditable, repeatable commercial returns. Capital can manufacture scale. It cannot manufacture proof of product-market fit.
Related Coverage:
Unitree’s Falling Floor Is Becoming a Ceiling for China’s Robot Startups
AgiBot Overtakes Unitree in H1 Humanoid Robot Shipments, But the Lead Remains Fragile