[Deep Analysis] Unitree's Market Capitalisation Has Fallen Below RMB 200 Billion

Prologue: RMB 240 Billion in Eleven Trading Days

• Part One · Longitudinal Analysis: A Decade of Making Robots Cheap
• Origins: The Electric-Drive Obsession of a "Lopsided" Student (1990–2016)
• The Darkest Hours and "Chinese-Style Cost Reduction" (2016–2020)
• Breaking Out: From the Spring Festival Gala to the Era of the Robot Dog (2020–2023)
• The Leap: Betting on Humanoids and the RMB 99,000 G1 (2023–2024)
• Reaching the Summit: From RMB 12.7 Billion to IPO Acceptance (2025)
• The IPO: Cleared in 73 Days, Amid a Trail of Troubles (H1 2026)
• The Gong and the Bubble: 19 August
• The Fall: From RMB 444.9 Billion to Below RMB 200 Billion
• Longitudinal Summary: What Unitree's Decade Proved, and What It Did Not
• Part Two · Cross-Sectional Analysis: A Company That Reigns in "Bodies" Encounters a Race for "Brains"
• Determining the Scenario: Full Competition (Scenario C)
• AgiBot: The Rival Most Like Unitree, and Most Unlike It
• UBTech: The Public Market's "Reference Case" for Unitree
• Galbot: The Chinese Standard-Bearer for a "Brains-First" Route
• Figure AI: The Western Template That Uses "Brains" to Set the Valuation Ceiling
• Tesla Optimus: The Biggest Variable in the Endgame of Scale
• The Remaining Players at a Glance
• Cross-Sectional Comparison Table
• The User Perspective: What Are People Actually Buying Unitree For?
• Niche Analysis: Where Unitree Sits in the Sector Map
• Trend Assessment: Where Unitree Goes from Here
• Part Three · Longitudinal–Cross-Sectional Convergence: What Does RMB 200 Billion Actually Mean?
• The Destiny of the Longitudinal, the Mirror of the Cross-Sectional
• Three Questions That Must Be Answered
• Integrated Judgement: My View
• Appendices
• Appendix A: Unitree Key Timeline
• Appendix B: Principal Data Sources
• Appendix C: Disclaimer

Prologue: RMB 240 Billion in Eleven Trading Days

At 9:30 a.m. on 19 August 2026, on the STAR Market of the Shanghai Stock Exchange, ticker 688836 — "Unitree Technology-W" — began trading. The moment the market opened, the screen flashed a number: RMB 1,100 per share. Against an offer price of RMB 150.80, that was a first-day gain of 629.44 per cent. At 404 million shares outstanding, the company's market capitalisation leapt instantaneously to RMB 444.9 billion. On the trading floor, institutional investors forwarded screenshots to one another with abandon — "winning one lot (500 shares) nets RMB 470,000." It was the most expensive IPO subscription event of the A-share year, and the coronation of the "first listed humanoid robot company." Founder Wang Xingxing stood on stage, composed. A reporter later noted that his expression as he struck the gong was taut. Three weeks later, the market wrote a different ending to the story. On 2 September, the share price fell below RMB 550 for the first time — halving from its intraday peak on the debut day. On 11 September, the price closed at RMB 477.12, and total market capitalisation stood at RMB 192.978 billion — formally breaching the RMB 200 billion mark, a cumulative decline of 55.45 per cent from the RMB 1,100 peak on 19 August, and an evaporation of roughly RMB 240 billion in value. Two hundred billion was wiped out in five trading days; the "halving" was completed in eleven. That is the origin of the statement "Unitree's market capitalisation has fallen below RMB 200 billion." But the number itself is not the interesting question. The real question is this: a company that in 2025 generated RMB 1.708 billion in revenue and RMB 288 million in net profit, ranked first globally in humanoid robot shipments, held the global number-one position in quadruped robots for several consecutive years, and was one of the very few integrated robotics manufacturers in the entire industry to be genuinely profitable — why did the market strip half its value within eleven trading days of listing? Did something go wrong inside the company, or had the market simply been wildly wrong beforehand? Or are both propositions true at once? To answer that, looking at the few candlesticks from August 2026 is insufficient. You need to stretch the timeline back a decade — to a boy in Yuyao, Ningbo, who was poor at English, and who built his first bipedal robot for RMB 200. And you need to widen the lens — to see how the company's rivals are faring at this very moment. What follows traces two lines of inquiry: the longitudinal and the cross-sectional.

Part One · Longitudinal Analysis: A Decade of Making Robots Cheap

1. Origins: The Electric-Drive Obsession of a "Lopsided" Student (1990–2016)

1.1 A boy from Yuyao, and a technology path the mainstream had abandoned Wang Xingxing was born in 1990 in Yuyao, Ningbo, Zhejiang Province. He was not the "well-rounded high-achiever" type. By his own account, he sat the English examination 300 times and passed only three. When he applied for postgraduate study at Zhejiang University, he was rejected on the strength of a single failing subject grade in English and was reassigned to Shanghai University. He would later describe the episode as "a blessing in disguise." He knew his strengths early. In junior secondary school he built a miniature turbojet engine; in senior secondary school he produced a chemical rechargeable battery; at university, using roughly RMB 200 of materials and a dozen 9-gram servo motors, he assembled a desktop bipedal humanoid robot. In 2009 he enrolled in the School of Mechanical Engineering and Automation at Zhejiang Sci-Tech University — a non-985, non-211 institution, reading mechanical engineering. By the standards of the time, that path was almost destined to be unremarkable. The turning point came during his master's degree. From 2014 to 2016, Wang Xingxing was a master's student in mechanical engineering at Shanghai University. During this period he independently developed a small quadruped robot called XDog, and made a decision that ran firmly against the prevailing orthodoxy: he chose pure electric drive, not hydraulic drive. The significance of that decision needs to be placed in its era. Around 2015, the global benchmark for quadruped robotics was Boston Dynamics. Its BigDog and Spot platforms used hydraulics — extraordinarily capable, able to somersault and sprint, but expensive and extremely difficult to manufacture at scale. A hydraulic quadruped could cost several hundred thousand US dollars. The industry's default assumption was: high-performance quadruped equals hydraulics. Wang Xingxing had no money. He had no laboratory budget, no supply chain, and nothing but himself, a dormitory, and cheap off-the-shelf components. He made a pragmatic judgement: I cannot afford hydraulics, so I will work around them. He selected outrunner brushless motors, rebuilt the structure himself, wrote the control algorithms himself, and eventually compressed the total hardware cost to under RMB 20,000. To finish the machine he voluntarily delayed his graduation by six months. In 2015, XDog won second prize in the Xingchuangshi Intelligent Entrepreneurship Competition, carrying a cash award of RMB 80,000 — his first pot of money. He posted the test videos online, but within China they made almost no splash. Chinese investors at the time were watching community group buying, O2O and live streaming. Nobody cared about a RMB 20,000 robot dog. 1.2 Two months at DJI, and a RMB 2 million cheque In June 2016, Wang Xingxing graduated with his master's degree and accepted an offer from DJI, moving south to Shenzhen as an engineer in motor control. His original intention was to leverage a major company's resources to advance his legged-robot project. His entire tenure in that role lasted a little over two months. Within those two months, events escaped his control. XDog's video was covered by IEEE Spectrum, shared by overseas YouTube creators, and went viral in the international robotics community. Overseas research institutions proactively sent purchase enquiries, and an investor named Yin Fangming — who had previously worked at MediaTek and Qihoo 360 — approached him and offered RMB 2 million. As to why he left DJI, he explained it publicly during the IPO roadshow: first, a mismatch of track — DJI's core business was consumer drones, and in 2016 quadrupeds were an extremely niche frontier direction with no clear path to commercialisation, so they could not command core resources; second, an external window had suddenly opened — the video had broken out, the investment had arrived, and prospective customers existed; third, a personal technological ambition — he wanted to see the full arc from prototype and self-developed motors through to product commercialisation, on his own terms. On 26 August 2016, Hangzhou Unitree Technology Co., Ltd. was incorporated. Registered capital was RMB 100,000; the office was 50 square metres; the founding team numbered three or four people; the starting capital was that RMB 2 million angel cheque. Yin Fangming received 15 per cent of the equity for his RMB 2 million, implying a post-money valuation of roughly RMB 13.33 million. This early history must be understood in full, because everything that followed depends on it. From day one, Unitree was not a company that told stories; it was a company that had been forced into existence by cost constraints. All of its technology choices — in-house motors, in-house reducers, in-house controllers, in-house LiDAR — were initially not "strategy" but poverty. It had to build these things itself because it could not afford to buy them. And that forced decision, a decade later, became its most durable moat: more than 90 per cent of core components are developed and manufactured in-house, delivering cost control that competitors cannot replicate. By the same token, that RMB 13.33 million valuation planted a seed with a ten-year fuse. From RMB 13.33 million to an offer-price market capitalisation of RMB 60.9 billion is a valuation leap of some four to five thousand times. Every layer of premium in the private market must eventually find a buyer in the public market. That is a story for later — but the seed was sown in 2016.

2. The Darkest Hours and "Chinese-Style Cost Reduction" (2016–2020)

2.1 Four years on a cold bench The first few years after incorporation were grim. Robotics was then a niche so obscure as to be nearly ignored. Consumer quadruped robots were broadly dismissed by mainstream institutions — "what would anyone buy this for?" was the standard objection. At the company's most difficult moment, its bank balance fell to just over RMB 100,000, and it was on the verge of being unable to pay salaries. Wang Xingxing suspended his own salary and kept the company alive on a trickle of orders. The product line nevertheless advanced, one machine at a time: • September 2017: Laikago launched (named after Laika, the Soviet space dog). This was Unitree's first commercialised quadruped, aimed at the research market. Wang Xingxing redesigned the motors, motor drives, main controller, mechanical structure and control system from scratch; compared with the XDog prototype, Laikago had more sensible proportions, stronger power and greater stability. • 2019: Aliengo launched. Positioned as an industrial functional quadruped, it used a newly designed powertrain with a lighter, more integrated body. This machine could perform a backflip — at the time, world-class performance among comparable international products. It was also in this year that Unitree began to build a reputation in the overseas research market. 2.2 The critical decision: why electric drive and full-stack self-development Return to the decision logic. Between 2016 and 2019, another path lay open to Unitree: become an integrator — buy off-the-shelf motors, reducers and sensors, and focus on assembly and algorithms. It was lighter, faster and easier to finance, and it was the choice most robotics companies made. Unitree did not take it. The reason was very practical: it could not afford to, and it could not source suitable parts. The joint motors that a high-performance quadruped requires simply did not exist as mature, inexpensive products; industrial motors were costly, and hydraulics were worse. Wang Xingxing had no choice but to build them himself. That "forced" decision accidentally established a self-reinforcing industrial logic: In-house core components → substantial cost reduction → lower product prices → larger shipment volumes → economies of scale further compress costs → still lower prices to expand share again

A decade later, a Huaxin Securities research note summarised Unitree's advantage thus: "Unitree's core component self-development rate exceeds 90 per cent; the high degree of hardware self-manufacture confers exceptional cost control, enabling the company to maintain a high level of profitability while keeping unit prices low." In 2025, the gross margin on Unitree's humanoid robot products reached 63.2 per cent. 2.3 Capital begins to enter — still against consensus Before 2019, robotics investment was not a mainstream activity. Those willing to bet during the "quiet period" were a small number of contrarian institutions: • December 2019: Ningbo HongShan (Sequoia China) led a Pre-A round of RMB 15 million at a post-money valuation of only about RMB 150 million. This was the first formal entry by a top-tier venture capital firm. Sequoia China subsequently invested a cumulative RMB 102 million and held 25,899,600 shares after the IPO, representing a paper return of more than RMB 2.5 billion. • 2020: Vertex Growth led the Pre-A+ round — an exceptionally rare instance of a round dominated by a single institution with no competing bidders. At the time, capital was concentrated in consumer internet, community group buying and online education; the quadruped-plus-humanoid thesis sat at the edge of institutional attention. • Other participants in this period included Anchuang Technology, Bianliang Capital (investing through Jisi Investment) and Dexun Investment. Worth noting is DJI's missed opportunity. As Wang Xingxing's former employer, DJI planned in 2018 to participate in Unitree's second capital increase through its own fund, committing RMB 10,128,600 at a post-money valuation of only RMB 60 million. The administrative change of registration was even completed. But DJI ultimately did not actually invest. Had the investment proceeded and been held through to the IPO, it would have been worth close to RMB 3 billion at the later market capitalisation — making DJI one of the largest external winners among Unitree's investors. It is a footnote about "missing out" that is still recounted today.

3. Breaking Out: From the Spring Festival Gala to the Era of the Robot Dog (2020–2023)

3.1 The A1 and the Spring Festival Gala: the first taste of mass exposure January 2020: Unitree launched the A1 at CES in Las Vegas — an ultra-high-value educational and recreational quadruped, compact and agile, with a maximum sustained outdoor running speed of 3.3 m/s. February 2021: Twenty-four of Unitree's A1 robots appeared on China Central Television's Spring Festival Gala for the Year of the Ox, in the guise of "little oxen Benben", performing Niu Qilai alongside Andy Lau, Wang Yibo and Guan Xiaotong. This was Unitree's first genuine breakout — from research equipment to a "technological symbol" in the public imagination. June 2021: The Go1 launched. This was Unitree's first quadruped truly aimed at the mass consumer market, starting at roughly RMB 20,000 with a maximum running speed of 4.7 m/s and an Intelligent Side-follow System (ISS). February 2022: 109 Go1 robots formed the word "Winter Olympics" at the Beijing Winter Olympics opening ceremony, appearing as representatives of the host city of the upcoming Asian Games, Hangzhou. February 2023: a Unitree Go1 danced for the American singer Jason Derulo during the Super Bowl pre-game show. That same year, at the Hangzhou Asian Games, Unitree robot dogs transported discuses and javelins — the first use of robots for such tasks in the history of global sporting events — earning the nickname "the show-offs"; at the Asian Para Games torch relay, the industrial-grade quadruped B1 debuted as an intelligent guide dog, providing technological support to disabled torchbearers. 3.2 Driving the quadruped down to RMB 10,000 The timeline now accelerates: • 2022: the industrial-grade quadruped B1 launched, deployed in power-grid inspection, firefighting and rescue. • 12 July 2023: the Go2 launched, starting at RMB 9,997 — the first time an "embodied-intelligence quadruped" had been pushed below RMB 10,000, with a 4D ultra-wide-angle LiDAR as standard and support for large-model interaction. • 3 November 2023: the industrial quadruped B2 launched, with a speed of 6 m/s, a sustained payload of 40 kg, five hours of endurance, and claimed comprehensive performance two to three times that of existing quadrupeds worldwide. By this stage, Unitree had effectively defined the quadruped category: it was not the highest-performing, but it had turned "high performance" into "affordable". Between 2023 and 2025, Unitree's cumulative quadruped sales exceeded 33,000 units, ranking first globally for several consecutive years. The most important change of this period, however, occurred at the industry level: in 2022, Tesla's Optimus and Figure AI both accelerated their humanoid programmes. "Humanoid" — until then a science-fiction concept — was placed seriously on the capital markets' agenda for the first time. And Unitree, for the first time, was pushed into the position of a "core robotics asset".

4. The Leap: Betting on Humanoids and the RMB 99,000 G1 (2023–2024)

4.1 The H1: China's first full-size humanoid that could run August 2023: Unitree unveiled its first general-purpose humanoid, the H1, described as "China's first full-size general-purpose humanoid robot that can run", with high dynamic performance among comparable specifications globally. It was a signal: Unitree's technology stack had migrated from four legs to two. Why was the migration not abrupt? Because quadrupeds and bipeds share the overwhelming majority of their underlying capabilities — motors, reducers, controllers, motion-control algorithms, reinforcement-learning frameworks. Unitree was not building a humanoid from scratch; it was reusing its core technology in a different body form. In Wang Xingxing's words, this was the "natural extension of the motion-control algorithms and self-developed core components from quadruped robots". 4.2 The G1: RMB 99,000, the world's first affordably mass-produced humanoid 13 May 2024: the G1 launched, priced from RMB 99,000. Unitree called it a "humanoid agent, an AI avatar", roughly 127 cm tall and 35 kg, with 23 to 43 joints. That price was startling at the time. UBTech's Walker series was priced in the hundreds of thousands of renminbi; Figure and Tesla had target prices of USD 20,000 to 30,000 (roughly RMB 140,000 to 210,000) and had not yet reached volume production. The G1 brought the price threshold for a general-purpose humanoid below RMB 100,000 for the first time. In late 2024, a test video of the B2-W (a wheeled-leg upgrade of the B2) went viral worldwide — it could climb hills and ford water, and carry a person across rugged terrain. The video elevated Unitree's overseas profile further still, and "the Chinese robot dog" became a global talking point. 4.3 The opening bars of capital frenzy Capital also began to accelerate: • April 2022: Series B+ (Shenzhen Capital Group, China Internet Investment Fund, Shunwei Capital, Rongyi, Dunhong and others) at a post-money valuation of RMB 1.12 billion. • February 2024: Series B++ of nearly RMB 1 billion (Meituan's strategic investment arm, CITIC Jinshi, Source Code Capital and others) at a post-money valuation of RMB 3.1 billion. • September 2024: Series B+++ (Sequoia China, Meituan, Shanghai Science and Technology Innovation Fund, Zhongguancun Science City and others) at a post-money valuation of RMB 3.8 billion. Note the pace: from RMB 1.1 billion in 2022 to RMB 3.8 billion in 2024 — less than a fourfold increase in just over two years. By the standards of the robotics sector at the time, that was, if anything, restrained. The real explosion came in 2025.

5. Reaching the Summit: From RMB 12.7 Billion to IPO Acceptance (2025)

5.1 The Year of the Snake Gala: the phenomenon of YangBot 28 January 2025 (Chinese New Year's Eve, Year of the Snake): sixteen (some reports say twenty) Unitree H1 humanoids performed the creative fusion dance 《秧BOT》 (YangBot) on CCTV's Spring Festival Gala, directed by Zhang Yimou. The robots spun handkerchiefs and tossed and caught them with a smoothness bordering on the uncanny. After that night, Unitree ceased to be a company known only within technology circles and became a national talking point. Brand searches, product enquiries and sales demand all surged. One detail is rarely emphasised: to enable the handkerchief-spinning motion, Unitree's engineers added three additional joints to each arm (on top of the original nineteen). That very engineering showmanship — tweaking hardware for stage effect — also laid the groundwork for the later criticism that the company was "strong on performance, weak on utility". 5.2 The humanoid sales explosion of 2025 After the Gala, Unitree entered a phase in which everything seemed to be taking off: • February 2025: humanoid robots went on sale on JD.com; the most expensive model (the H1) was priced at RMB 650,000 and sold out quickly. Unitree stated that "any ordinary user can buy one". • 25 May 2025: the world's first humanoid robot combat tournament based on the G1 — robots trading blows, and standing back up after falling. • 25 July 2025: the third humanoid, the R1, launched from RMB 39,900, weighing about 25 kg with 26 degrees of freedom, aimed at education and developers. • 5 August 2025: the new quadruped A2 "Star Cheetah" launched, weighing about 37 kg. • 14–17 August 2025: at the inaugural World Humanoid Robot Games in Beijing, Unitree won four gold medals — in the 1,500 m, 400 m, 100 m obstacle race and 4×100 m relay — topping both the gold-medal and overall medal tables. The H1 reached a top speed of 5+ m/s. • 20 October 2025: the full-size humanoid H2 launched, at 180 cm and 70 kg, with a bionic face and dual-eye cameras. That year, Unitree topped the global rankings in both quadruped and humanoid shipments. In 2025 it shipped more than 5,500 humanoid units (pure humanoids, excluding wheeled dual-arm robots), a global market share of 32.4 per cent and the number-one position worldwide; cumulative quadruped sales for 2023–2025 exceeded 33,000 units, again first globally for several consecutive years. 5.3 The twin explosion of performance and capital The financials exploded in tandem: Year Revenue Net profit attributable to shareholders Net profit excluding non-recurring items Gross margin (main business) 2022 RMB 123 million — −RMB 8.07 million 44.18% 2023 RMB 159 million −RMB 11.1451 million −RMB 18.02 million 44.22% 2024 RMB 392 million RMB 94.5018 million RMB 77.5 million 56.41% 2025 RMB 1.708 billion (+335.36%) RMB 288 million Approx. RMB 590–600 million (+674.29%) 60.13%–60.44%

2025 was Unitree's pivotal year: it became one of the very few — arguably the first — humanoid robot manufacturers globally to achieve both scale shipments and full-year profitability. Note the anomaly that net profit excluding non-recurring items (approximately RMB 600 million) exceeds net profit attributable to shareholders (RMB 288 million) — implying roughly RMB 300 million of non-recurring losses in 2025 (widely understood to relate to share-based payments and similar items; the precise treatment should be read against the company's own disclosure). In the context of the time, this set of figures was cited repeatedly, because it constituted the most compelling available answer to the question of whether humanoid robots can make money. Capital followed suit: • June 2025: Series C of approximately RMB 700 million at a post-money valuation of RMB 12.7 billion (some sources indicate RMB 13.36 billion), co-led by China Mobile's fund, Tencent, Alibaba, Ant Group and Geely Capital, with the overwhelming majority of existing shareholders participating. • 18 July 2025: filing for IPO coaching with the Zhejiang office of the CSRC, with CITIC Securities as sponsor. • November 2025: IPO coaching status updated to "coaching completed". 5.4 Wang Xingxing's year of apotheosis In 2025, Wang Xingxing himself completed the transition from entrepreneur to "figure of the age": on 30 April he received the China Youth May Fourth Medal; in June he was appointed to the Hong Kong Chief Executive's Advisory Group; in July he was named 2025 AI Person of the Year and appeared on the cover of TIME magazine — the previous Chinese entrepreneur to do so was Jack Ma, eight years earlier. Yet his attitude towards the frenzy was clear-eyed from the outset. In the TIME interview he said: "Because the hype has been excessive, it has brought considerable pressure on both the company and the entire industry. Everyone expects your technology to leap forward at extreme speed. But the fact is that breakthroughs in hard-core technology take time." That remark would later be treated by the market as prophecy.

6. The IPO: Cleared in 73 Days, Amid a Trail of Troubles (H1 2026)

6.1 A green light all the way In 2026, Unitree's listing process was nothing short of lightspeed: • 20 March 2026: the Shanghai Stock Exchange accepted Unitree's STAR Market IPO application, seeking to raise RMB 4.202 billion. From acceptance to approval took only 73 days, a record for the fastest IPO review that year. • 1 June 2026: approved by the SSE Listing Committee. • 2 July 2026: registration approved by the CSRC. • 5 August 2026: preliminary price consultation; 6 August: the offer price was set at RMB 150.80 per share, implying a market capitalisation of approximately RMB 60.9 billion. The use of proceeds also revealed a strategic pivot: of the RMB 4.202 billion, RMB 2.022 billion — nearly half — was allocated to the "intelligent robot model R&D project", namely the "brain" and "cerebellum" core technologies. That scale is an order of magnitude larger than the approximately RMB 265 million the company had spent on R&D cumulatively over the preceding three years. 6.2 Three pre-IPO "complications" Unitree's IPO was nonetheless not clean. First, a patent shakedown. Beginning in July 2025, Hangzhou Luweimei Daily Chemicals Co., Ltd. — a household chemicals company with no connection whatsoever to robotics — filed a series of patent infringement suits against Unitree over several of its best-selling quadruped products, invoking a single invention patent titled "An Electronic Dog". The conduct was theatrical: it sued only five days after acquiring the patent; at first instance it claimed Unitree had profited by tens of millions of renminbi, yet sought only RMB 500 in damages; at second instance it first asserted RMB 80 million, then the day after the hearing filed in writing to reduce the claim back to RMB 500. In February 2026 the Supreme People's Court dismissed its appeal in full, expressly criticising its conduct as "both carefully calculated and persistently inconsistent" and "in violation of the principle of good faith"; in a separate case the Hangzhou Intermediate People's Court found it liable for malicious litigation and ordered it to pay RMB 80,000 towards Unitree's costs; in March, the patent in question was invalidated. Unitree won — but the litigation consumed substantial management attention. Second, a spot inspection. On 1 April 2026, the Securities Association of China published the second batch of listed-company applicants selected for on-site inspection — and Unitree was among them. Such inspections cover financial veracity, internal controls and compliance across multiple dimensions, and during a sensitive IPO period directly affect the review timetable and market confidence. In the same period, the market scrutinised the compliance of the company's revenue recognition on related-party transactions — the prospectus showed that in the first three quarters of 2025, more than 70 per cent of humanoid robot revenue came from research and education procurement, and the verifiability and sustainability of that revenue were repeatedly questioned. Third, labour disputes came to light. A list of 32 items of evidence submitted by a Unitree employee in a second-instance labour dispute circulated online, concerning "17 consecutive days at an average of 12.2 hours per day", the attribution of responsibility for a fire incident, and an extreme "suicide-threat protest" followed by detention. Separately, on 1 September 2026, several current and former employees publicly alleged that the company's pay was significantly below that of peer embodied-AI start-ups, that any reimbursement above RMB 100 required Wang Xingxing's personal approval, that the reward-and-penalty system was primarily punitive, and that 2025–2026 saw the highest attrition rate among core staff. Unitree responded that "much of the content is untrue and should not be taken at face value". These are one-sided allegations, not judicial findings — but they shape the public's impression of the company's human dimension. 6.3 Technical highs, and America closing the door In the first half of 2026, Unitree continued to post technical highlights: • 11 April: the H1 ran approximately 10 m/s in a 100 m test, setting a world record for humanoid running speed and approaching the limits of human sprinting. • 16 April: at the Beijing humanoid robot marathon qualifier, the H1 autonomously completed a 1.9 km multi-curve course in 4 minutes 13 seconds, breaking the human world record for 1,500 m on a proportional basis. • 30 April: launched dual-arm humanoids (the R1-A5/R1-A7 series) from RMB 26,900. • 12 May: launched the GD01, the world's first mass-produced manned transforming mecha, about 3 metres tall, weighing approximately 500 kg when manned, priced from RMB 3.9 million. • 1 June: launched the H2 Plus, a compute-enhanced version developed jointly with NVIDIA. • 17 August: unveiled a new robot, the "Superman", capable of a standing jump of about 2 metres and a top running speed of 12+ m/s. But the US market began closing in the same period: • 8 June: the US Department of Defense added 65 entities to its Section 1260H "Chinese military companies" list, including Unitree. The consequences arrive in two steps: from 30 June 2026 the DoD may not contract directly with listed companies; from 30 June 2027 it may not procure goods or services containing their components. • 28 July: the US Federal Communications Commission (FCC) added "advanced robotic devices produced abroad" (covering humanoid and quadruped robots) to its Covered List. The new rule applies only to new models — Unitree's existing models (G1/H2/R1; Go2/B2/A2) obtained FCC certification before the policy took effect and may continue to be sold in the US, but future models will be unable to enter the US market unless granted a conditional approval. As of early September 2026, the US Department of Defense had granted no conditional approvals to any foreign-made robotic device. • The prospectus disclosed that between 2023 and 2025, Unitree's overseas revenue exceeded 40 per cent each year, with the United States accounting for 18.39 per cent, 19.54 per cent and 13.30 per cent respectively. China's Ministry of Commerce responded that the US was "generalising the concept of national security" and "using administrative power to artificially interfere in normal commercial transactions and trade between enterprises", describing it as "a classic act of market distortion and unilateral bullying". For a company deriving more than 40 per cent of revenue from overseas, this is a sword suspended overhead.

7. The Gong and the Bubble: 19 August

7.1 The madness of RMB 444.9 billion On 19 August 2026, Unitree listed on the STAR Market. The offer price was RMB 150.80, implying an offer market capitalisation of RMB 60.9 billion and an offer price-to-earnings ratio of 219.23 times. The stock opened at RMB 1,100, a gain of 629.44 per cent, taking market capitalisation to RMB 444.9 billion. It then dived, at one point falling to RMB 800.08 intraday before closing at RMB 845 — a retreat of roughly 23 per cent from the opening price. Why could a debut day that opened high and closed low still produce a market capitalisation of RMB 444.9 billion? The answer lies in the shareholding structure. The prospectus showed that the public offering comprised 40,446,434 shares, 10 per cent of total post-issue share capital; of these, 20,380,700 shares were allocated off-line without a lock-up and 9,707,000 shares on-line. The shares genuinely free to trade on the first day amounted to only 7.44 per cent of total share capital. Against such a tiny free float, first-day turnover exceeded 85 per cent — almost the entire tradeable supply changed hands. A modest amount of capital was sufficient to drive an enormous total market capitalisation to an absurd level, and then to collapse it rapidly. This is a classic defect of A-share "new stock speculation"; Unitree's extremity lay in the combination of a very small free float and very high attention. 7.2 The offer price itself was already expensive One misconception must be corrected: Unitree was not "a poor-quality company that was hyped up". Its offer price was already expensive. After assessing fair investment value, peer valuations, secondary-market valuations for the industry, and other factors, the issuer (with lead underwriter CITIC Securities) set the price at RMB 150.80, implying 219 times earnings. The static price-to-earnings ratio for the general equipment manufacturing sector to which the company belongs was only 38.56 times, and for the Hong Kong-listed comparable UBTech it was approximately 19.37 times. Even on the basis of net profit excluding non-recurring items of RMB 590 million, the offer P/E was around 100 times. GuoXin Securities noted at the time that "the core assumption embedded in a P/E above 219 times is that Unitree will need to grow at a rate far exceeding the industry average for several years in order to digest the current valuation." Institutional expectations for the post-listing market capitalisation varied widely: • CITIC Securities (lead underwriter): a fair value range of RMB 50.6 billion to 55.9 billion; • CCB International: a forecast market capitalisation of RMB 109 billion; • Nomura: a target price of RMB 370, implying approximately RMB 150 billion. The secondary market opened at RMB 444.9 billion — nearly three times the most optimistic institutional forecast. Hence the "black humour": one investor calculated that "the company spends roughly RMB 100 million a year on R&D, and it opened with a market capitalisation of over RMB 400 billion, on annual revenue of still only RMB 1.6 billion."

8. The Fall: From RMB 444.9 Billion to Below RMB 200 Billion

8.1 Wang Xingxing pours cold water the day after listing On 20 August, the day after listing, Wang Xingxing appeared at the "Application-Led" plenary forum of the 2026 World Robot Conference (WRC 2026) and delivered a speech titled From Exhibits to Products: The Next Decade of the Humanoid Robot Industry. It was his first public appearance after the listing. He spoke exclusively about robots, never mentioning the listing, the share price or his personal wealth, and he spoke very quickly. Then he did something that struck the capital markets as highly unusual — he deliberately cooled the room: • "Robots can already perform some simple assembly, but their efficiency is lower than that of humans, and every new task requires retraining. This is the common bottleneck the whole world faces." • "In a fixed scenario with sufficient data collection and training, the success rate can approach 100 per cent; but if the objects or environment change even slightly, the success rate falls off very sharply." • He revised his expectation for the humanoid robot "ChatGPT moment" from "at the soonest one to two years, at the latest three to five years" to "at the soonest two to three years, at the latest five to ten years." That day, Unitree's share price continued to slide; RMB 63.9 billion of market capitalisation evaporated in two days. The market read the remarks as confirmation that the founder "knew all along that it would fall". Objectively, Wang Xingxing was telling the industry's truth — he had said the same things before the listing. But at a moment when the valuation had just been pushed to RMB 444.9 billion, the founder's honesty was equivalent to the bubble confirming itself. 8.2 Eleven trading days of value reversion What followed was a textbook sequence: • 2 September: the share price fell below RMB 550 for the first time, halving from the RMB 1,100 peak, with more than RMB 228 billion of market capitalisation destroyed; • 11 September: the share price fell to RMB 477.12 and total market capitalisation stood at RMB 192.978 billion — still about 217 per cent above the offer market capitalisation of RMB 60.9 billion, but down 55.45 per cent from the first-day opening price. Several forces combined to drive the decline: First, an abrupt slowdown in growth. The listing announcement disclosed that H1 2026 revenue was RMB 1.152 billion, up 48.54 per cent year on year; net profit attributable to shareholders was RMB 274 million; and net profit excluding non-recurring items was RMB 244 million, down 19.34 per cent. Compare that with revenue growth of 335.36 per cent for the whole of 2025 — growth had dropped by an order of magnitude. And in Q1 2026 specifically: revenue of RMB 423 million (+68.49 per cent), net profit attributable to shareholders of RMB 50.0138 million (−47.69 per cent) and net profit excluding non-recurring items of RMB 40.2536 million (−52.55 per cent). The company explained in its announcement that the decline in adjusted net profit was primarily due to a substantial year-on-year increase in period expenses including R&D and selling expenses. What a growth story fears most is not an absence of growth but a step down in the growth rate — the market will pay 100 times earnings for 200 per cent compound growth, but it will not pay several hundred times earnings for 48 per cent growth. Second, the gulf between valuation and fundamentals. On 2025 net profit attributable to shareholders of RMB 288 million, a market capitalisation of RMB 197.8 billion implies a P/E ratio still in the hundreds, with a payback period measured in millennia; even on adjusted net profit of RMB 600 million, the P/E remained above 300 times. In the same period, UBTech's market capitalisation was about HKD 42 billion and Shangwei New Materials (acquired by AgiBot) about RMB 56.5 billion — Unitree's market capitalisation was at one point a multiple of theirs, yet in H1 2026 its revenue (RMB 1.152 billion) was actually lower than UBTech's (RMB 1.269 billion). That comparison was cited repeatedly by the media and became the most vivid evidence of a bubble. Third, the criticism that Unitree is "heavy on hardware, light on brains". The industry widely regards Unitree as a "lopsided" player in robotics — first-tier in motion control, but with clear shortcomings in embodied large models, environmental cognition and decision-making. The prospectus itself concedes that its "cerebellum is developed while its brain is lacking". The data corroborates this: in 2025 roughly 70 per cent of revenue came from research and education, with industrial applications accounting for only about 9 per cent; R&D spending in 2025 was RMB 145 million, only 8.53 per cent of revenue, far below UBTech's 25.4 per cent and below the roughly 11 per cent average for STAR Market companies. The SSE's IPO enquiry specifically addressed "the continued decline in the R&D expense ratio, the fact that invention patents represent less than 8 per cent of the total, and the sustainability of earnings growth". The chairman of TSMC even publicly questioned whether Unitree's robots "just keep jumping around and bouncing — useless, just for show". Fourth, the sword of Damocles that is the lock-up expiry. Little is available to sell in the near term — the off-line subscription portion locked up amounts to only about 10 per cent (approximately 2.269 million shares), due to unlock in February 2027, a limited quantity. The real sluice gate opens a year from now: on 19 August 2027, the 12-month lock-up for pre-IPO shareholders including Meituan, Sequoia and Jingwei will expire, releasing approximately 229 million shares, 56.6 per cent of total share capital. That will be the largest tranche of selling pressure since listing. Every rational investor is asking: a year from now, who will absorb that 56.6 per cent? Fifth, the arithmetic of founder and shareholder paper wealth. One calculation that circulated on Xueqiu was particularly pointed: Wang Xingxing founded the company in 2016 at a cost of roughly RMB 1 per share; the angel investor Yin Fangming received 15 per cent for RMB 2 million at RMB 0.15–0.20 per share; Sequoia's first round in 2019 was at a post-money valuation of RMB 150 million, or RMB 3–5 per share; Meituan entered in the 2024 Series B2 at a post-money valuation of RMB 3.1 billion; Tencent, Alibaba and China Mobile subscribed in the 2025 Series C at RMB 46.26 per share; and the IPO offer price was RMB 150.80. From RMB 1 to RMB 150, every layer of premium ultimately lands on the secondary market — retail investors buy at the top, while founding shareholders on a RMB 1 cost base are still up more than 430 times. That perception — the same share, with costs and returns at opposite extremes — becomes the most violent selling pressure once sentiment reverses. And so: RMB 1,100 to RMB 477, in eleven trading days.

9. Longitudinal Summary: What Unitree's Decade Proved, and What It Did Not

Read from beginning to end, three things about this timeline are clear. First, Unitree's decade is a history of Chinese-style engineering cost reduction. It substituted electric drive for hydraulics, in-house development for procurement, and scale for premium pricing, driving quadrupeds from several hundred thousand US dollars to RMB 10,000, and humanoids from the million-renminbi bracket to below RMB 100,000, then to RMB 39,900 for the R1 and RMB 26,900 for the dual-arm model. It demonstrated that the "expense" of hardware is often a matter of path dependency rather than physical law. This is its genuine, undeniable value. Second, Unitree proved that hardware can be profitable, but not that embodied intelligence can be. It turned a profit in 2025, yet 70 per cent of revenue came from research and education and only 9 per cent from industry; its best-selling products were bought mainly for performance, display and research rather than "work". It has secured a ticket to "scaled production plus profitability", but the ticket to "commercial value" has not yet been punched. Wang Xingxing himself put it most plainly at WRC: efficiency is lower than humans', and generality is insufficient. Third, Unitree's valuation was never a company valuation; it was a sector-option valuation from the start. The 219-times offer P/E and the RMB 444.9 billion first-day market capitalisation were not pricing 2026 earnings but discounting the distant trillion-renminbi market for humanoid robots. When market sentiment shifted from "capturing the future" to "watching the present", a correction became inevitable — the RMB 200 billion figure does not reflect a change in Unitree's fundamentals but a change in the market's discount rate.

Part Two · Cross-Sectional Analysis: A Company That Reigns in "Bodies" Encounters a Race for "Brains"

10. Determining the Scenario: Full Competition (Scenario C)

Applying the analytical framework, we first determine the competitive scenario in which Unitree finds itself.

Conclusion: Scenario C (full competition). Humanoid robotics and embodied intelligence is an extraordinarily crowded field. As of 2026, China has more than 200 embodied-intelligence companies, of which more than 100 are humanoid robot companies; globally there are more than 50 humanoid robot start-ups. In China alone, financing in the embodied-intelligence sector exceeded RMB 93.5 billion in the first half of 2026 (some tallies exceed RMB 100 billion), roughly five times the prior-year level.

Accordingly, this report selects five of the most representative and comparable competitors for detailed analysis — AgiBot (the most direct head-to-head rival), UBTech (the public-market reference point), Galbot (the standard-bearer of the "brains-first" route), Figure AI (the Western paradigm), and Tesla Optimus (the scale paradigm) — with the remainder (Galaxea, LimX Dynamics, EngineAI, DEEP Robotics, Fourier Intelligence, Agility Robotics, Boston Dynamics and others) addressed briefly. It should be noted that Unitree's true competitors actually span two distinct groups. One comprises Chinese peers who likewise started from the "body" (AgiBot, UBTech); the other comprises American companies attempting to use "brains" to swallow the body from downstream (Figure, Tesla). This "two-front war" is the key to understanding Unitree's position.

11. AgiBot: The Rival Most Like Unitree, and Most Unlike It

11.1 What it has become AgiBot (Zhiyuan Innovation) was founded in February 2023 — a full seven years after Unitree — yet it has become Unitree's most direct rival. Its two leading figures are a study in contrast: Deng Taihua, a 1995 cohort alumnus of the University of Electronic Science and Technology of China, who spent more than two decades at Huawei, serving as president of the wireless network product line and of the computing product line and leading the construction of the Kunpeng and Ascend AI computing ecosystems; and Peng Zhihui, born in 1993, better known by his Bilibili handle "Zhihui Jun", who joined Huawei's "Genius Youth" programme in 2020 at its highest salary tier and worked on AI and Ascend chip development. His hard-core DIY robotics videos earned him 2.69 million followers and the nickname "the wild Iron Man". In February 2023 the two founded AgiBot. AgiBot's growth has been remarkable: from RMB 300,000 of revenue in 2023 to more than RMB 1 billion in 2025. On the Omdia measure it shipped about 5,100 units in 2025, first globally; on Counterpoint's measure, it shipped about 9,700 units in H1 2026, giving it more than 43 per cent of the global market and first place. On 30 March 2026, its 10,000th general-purpose embodied robot, the Expedition A3, rolled off the line; on 28 June, the 15,000th did. Its product matrix spans wheeled, semi-humanoid, humanoid and quadruped form factors: the Expedition A3 (full-size bipedal, 173 cm/55 kg, positioned for entertainment performance but already deployed for loading and unloading in 3C factories); the Lingxi X2 (high-mobility biped, winner of the 100 m obstacle race at the World Humanoid Robot Games); and the Genie G2 (a task-operations model deployed on 3C production lines). Customers include Longqi Technology, Joyson Electronics, BYD, Midea and JD Logistics. 11.2 Capital operations: far more aggressive than Unitree If Unitree "won its territory with products", AgiBot "leverages capital to expand the space of the possible". In July 2025, AgiBot acquired 63.62 per cent of the STAR Market-listed Shangwei New Materials for approximately RMB 2.1 billion — a company that had made wind turbine blade materials and had a market capitalisation of just RMB 3.1 billion. On news of the acquisition, Shangwei's share price rose more than 1,083 per cent in sixteen trading days, becoming the year's first "ten-bagger", and the stock was placed on the SSE's key-monitoring list, with market capitalisation peaking at RMB 56.5 billion. AgiBot deliberately structured the deal to avoid being classified as a backdoor listing (taking 29.99 per cent in stages to stay below the 30 per cent mandatory-offer threshold, and committing to no backdoor listing within 36 months), and in July 2026 confirmed the launch of a Hong Kong listing process targeting a valuation of HKD 40–50 billion. Combined with its A-share platform, AgiBot is attempting to build a dual "A+H" capital platform. Chairman Deng Taihua has set out an ambitious "358 blueprint": revenues of RMB 10 billion by 2027 and a scale of RMB 100 billion by 2030. 11.3 The real differences from Unitree On technology, the two are mirror images. Unitree attacks "brains" from the starting point of the "body" (chassis plus motion control); AgiBot attacks the "body" from the starting point of "brains" (embodied models plus a Huawei-lineage AI gene). AgiBot presents itself as a "platform" company, with multiple platforms under its umbrella — Lingxin, Lingchuang, Genie Studio and Link-U OS — pursuing an open-source ecosystem (the Lingxi X1 is the world's first fully open-source humanoid) and a spinoff model (the Lingjie Dian dexterous hand; AgiBot Kuotuo for quadrupeds). On commercialisation, AgiBot is more "saturating R&D plus aggressive expansion". Chairman Deng Taihua has stated explicitly that the company is in no hurry to achieve positive net profit at this stage and is "focused on the future". Unitree is the precise opposite — one of the few companies that insists on profitability. The divergence in user perception is also instructive. AgiBot has benefited from the "Zhihui Jun" traffic dividend and enjoys high public recognition, but scepticism about the share of "performance orders" has never gone away. The turning point came in 2026 when AgiBot conducted an eight-hour live broadcast from Longqi Technology's Nanchang factory, demonstrating in real time that the cost of humanoid loading and unloading had fallen below that of human labour, and announcing plans to scale from four units to around a hundred in the third quarter of 2026. This was the first Chinese company to prove — by live broadcast — that humanoid robots can genuinely work; from that angle, on the specific question of "demonstrating commercial value", AgiBot briefly ran ahead of Unitree. In a sentence: Unitree is the "motion prodigy and cost killer"; AgiBot is the "Huawei-lineage iron army and capital operator". They fight hand-to-hand on shipments, while AgiBot is more adept at using the A-share market as an amplifier for its story.

12. UBTech: The Public Market's "Reference Case" for Unitree

12.1 What it has become UBTech (9880.HK), founded in 2012, is one of the earliest Chinese humanoid robot companies, and listed on the Hong Kong Stock Exchange in December 2023 as the "first listed humanoid robot company in Hong Kong". Its H1 2026 results are the best available reference for understanding Unitree's valuation bubble: Metric UBTech H1 2026 Unitree H1 2026 Revenue RMB 1.269 billion (+104.2%) RMB 1.152 billion (+48.54%) Net profit / loss for the period Loss of RMB 339 million Profit of RMB 274 million Full-size humanoid revenue RMB 590 million (+1,445%) — Full-size humanoid units 921 (average price ~RMB 640,000) — Total humanoid units 16,123 — Market capitalisation (end-Aug 2026) ~HKD 42–46 billion Peak of RMB 444.9 billion

For the full year 2025, UBTech reported revenue of RMB 2.001 billion (+53.3 per cent) and a net loss of RMB 790 million, with cumulative losses over four years exceeding RMB 4.2 billion; R&D expenses were RMB 507 million and selling expenses RMB 471 million, together close to 50 per cent of revenue. Although gross margin on the humanoid business reached 54.6 per cent, the heavy investment consumed that margin, and net operating cash outflow in 2025 was RMB 784 million. UBTech's strategy is a three-pronged push across industry, commerce and the home: in industry, the Walker S bipedal series and Cruzr wheeled series (covering bin handling, loading and unloading, sorting, parts installation and quality inspection); in commerce, the Walker C1; and in the consumer home segment, the Wanda companion robot. In 2025 it secured approximately RMB 1.4 billion of orders (the largest publicly disclosed order book of any robotics company) and in H1 2026 signed more than RMB 50 million of overseas orders with customers in Europe, Japan and South Korea. CFO Zhang Ju has stated clearly that the original expectation of single-quarter break-even in Q4 2027 is unchanged, that the company is trying to bring it forward, and that single-quarter adjusted EBITDA in Q4 2026 "has a chance of turning positive". 12.2 The key differences from Unitree UBTech is on the traditional hard-tech path of heavy R&D, heavy investment and prolonged losses; Unitree is on the extreme hardware path of high gross margin, rapid profitability and light R&D. On the R&D expense ratio, UBTech is at 25.4 per cent against Unitree's 8.53 per cent — this pair of figures is the key to understanding the difference in temperament between the two companies. More important is the pricing divergence in the secondary market. When Unitree's market capitalisation reached RMB 444.9 billion, UBTech's was only about HKD 42 billion (roughly RMB 36 billion). Yet in H1 2026, UBTech's revenue was slightly higher than Unitree's. The market's logic is clear: Unitree is "the only profitable number-one integrated manufacturer", commanding a scarcity premium; UBTech is "a long-loss-making veteran", trading at a discount. Pushed to its extreme, however, that logic becomes "using a company earning RMB 20 million to justify a RMB 200 billion valuation gap" — which is the bubble. From the user perspective, UBTech's B2B reputation is relatively solid (Airbus, BYD, Foxconn and others), but the market remains dissatisfied with its persistent losses and the pace at which its story converts into results. Some media have described it as having "doubled revenue, but delivered only half the promise" — the other half being a return to profit. In a sentence: UBTech is Unitree's "previous station" — if Unitree cannot make up ground on "brains", it may well end up looking like UBTech: revenue growing, profit under pressure, and valuation reverting towards traditional manufacturing multiples.

13. Galbot: The Chinese Standard-Bearer for a "Brains-First" Route

13.1 What it has become Galbot was founded in 2023, originating from Professor Wang He's team at Peking University, and is currently China's highest-valued unlisted humanoid robot company. In March 2026 it announced the completion of a RMB 2.5 billion Series A+ round, with investors including the National AI Industry Fund (the third phase of the "Big Fund"), Sinopec, CITIC Group Investment Holdings, Bank of China and SAIC Group's financial arm — the first investment by the Big Fund in the embodied-intelligence sector, and at the time the largest single robotics financing event in China. The company's valuation exceeds RMB 20 billion (approximately USD 3 billion). On products, Galbot offers the industrial heavy-duty S1 (maximum dual-arm payload of 50 kg) and the home/service G1 (24 degrees of freedom). Its technological standard is an "embodied large model plus end-to-end training framework", achieving zero-shot task transfer in warehouse sorting and home-service scenarios. It was designated by China Media Group as the "official embodied large-model robot for the 2026 Spring Festival Gala". In April 2026, AgiBot upgraded from customer to strategic shareholder in Galbot. 13.2 The differences from Unitree Galbot is the Chinese exemplar of the "brains-first" route — its core assets are models and data, and hardware can be outsourced and integrated. This stands in sharp contrast to Unitree's full-stack, self-developed body. The bets differ. Unitree is betting on "whose robot is cheapest, who can mass-produce, who can ship"; Galbot is betting on "whose model generalises first, becomes general-purpose first, and can handle tasks it has never seen". Judging by Unitree's own allocation of IPO proceeds (RMB 2.022 billion to model R&D), Unitree has in effect conceded that "brains-first" is the direction of the endgame — it simply started later and with a thinner foundation. From the user perspective, Galbot's "national team" backing and Peking University pedigree bring a trust premium, but its commercial scale (cumulative shipments of about 1,200 units in 2025) is far smaller than Unitree's. Its true benchmark is not Unitree but Figure — a company that uses a large model to define the valuation ceiling.

14. Figure AI: The Western Template That Uses "Brains" to Set the Valuation Ceiling

14.1 What it has become Figure AI was founded in 2022 by Brett Adcock, a serial entrepreneur, and is headquartered in San Jose, California. Its most recent round valued the company at approximately USD 39 billion (about RMB 269.1 billion), with more than USD 1.9 billion raised cumulatively from investors including Microsoft, NVIDIA, OpenAI, Jeff Bezos and Intel Capital — a shareholder list that is itself a signal: capital believes that large models will swallow hardware. It is the only Western humanoid robot company with genuinely paying external customers in 2026: • BMW's Spartanburg plant: Figure robots logged more than 1,250 operational hours and contributed to the assembly of over 30,000 X3 vehicles; in March 2026 BMW expanded the collaboration to its Leipzig plant in Germany — a customer-initiated expansion is the hardest signal to fake in enterprise hardware; • May 2026: Figure completed a 200-hour warehouse livestream, during which five robots sorted 249,560 packages with, the company says, zero failures; • The BotQ factory: targeting annual capacity of 12,000 units, reportedly already producing Figure 03 at 55+ units per week. Technologically, Figure's core asset is Helix 02 — a vision-language-action (VLA) model announced in January 2026 that controls the entire body with a single neural network running on board, learning tasks by observation rather than explicit programming. The Figure 03 is equipped with fingertip tactile sensors sensitive to 3 grams of force, embedded palm cameras, wireless inductive charging and a soft textile exterior. 14.2 The differences from Unitree This is "software defining hardware" colliding with "hardware defining software". • Unitree shipped 5,500+ humanoids in 2025; Figure shipped only a few hundred — on volume, Unitree crushes Figure; • But Figure's valuation (about RMB 269.1 billion) far exceeds Unitree's pre-IPO valuation of RMB 15–18 billion — on the space of the possible, Figure crushes Unitree; • Figure tells a "software entry point" story built on the Helix 02 brain and real customers; Unitree tells a "hardware platform" story built on high gross margins and full-stack self-development. From the customer perspective, Figure's customers are top-tier manufacturers such as BMW, paying, repeat-ordering and expanding; Unitree's customers are largely universities and research institutes that "buy the machine to do research". That gap in customer quality is the root of the difference in valuation logic. In a sentence: Unitree sells robots; Figure sells the option that "robots will think" — and the market always prices options above hardware.

15. Tesla Optimus: The Biggest Variable in the Endgame of Scale

15.1 What it has become Tesla's Optimus is the mountain nobody can route around. Its logic is brutally simple: transplant the FSD vision neural network, the Dojo compute stack and vertically integrated manufacturing capability onto a humanoid body. • Gen 3 targets a price of USD 20,000–30,000 (roughly RMB 140,000–210,000), below one year's cost of human labour; • The Fremont plant began Gen 3 production in July–August 2026, with an annual capacity target of one million units; Nomura estimates 2026 output at around 25,000 units, while Tesla's own target is 5,000–10,000; • Tesla has committed USD 25 billion of capital expenditure for 2026; • But as of mid-2026 Optimus had no external paying customers, and the 1,000+ units were mainly "learning" inside Tesla's own factories — Musk himself conceding that they were not yet performing meaningful productive work. 15.2 The differences from Unitree Tesla and Unitree are in fact working the two ends of the same problem: both are pursuing "make robots cheap, make robots numerous". The difference is that Tesla uses the mass-production capability and captive scenarios of the automotive industry to amortise cost (it is its own first customer), whereas Unitree uses in-house supply chain and product iteration to amortise cost (selling to universities and early customers). Tesla is Unitree's greatest long-term anxiety: if Tesla truly brings Optimus to USD 20,000 and produces it by the million, then the premise that "robots are expensive" disappears, and Unitree's cost moat is levelled — because Tesla's costs are lower, its capacity greater, and its AI capability (FSD) stronger. But Tesla is also Unitree's greatest source of support: as long as Tesla's million-unit vision stands, the story of the entire humanoid sector does not break, and Unitree's logic of discounting a distant trillion-renminbi market continues to hold. In a sentence: Tesla is both Unitree's "ceiling-smasher" and its "story-keeper" — at once the greatest risk to Unitree's valuation and an important external reason why that valuation can remain in the hundreds of billions.

16. The Remaining Players at a Glance

Company Valuation / Market cap Key developments 2025–2026 Route characteristics Galaxea ~RMB 20 billion Series B of nearly RMB 1 billion in Feb 2026 and Series B+ of nearly RMB 2 billion in Apr, ~RMB 3 billion in total; preparing a Hong Kong listing to raise about USD 500 million General embodied intelligence "brain", world models; highest-valued private embodied-AI company LimX Dynamics ~RMB 10 billion USD 200 million (≈RMB 1.4 billion) Series B in Mar 2026, with JD.com, SAIC, NIO and Abu Dhabi's Stone Venture Full-stack body, cerebellum and embodied OS; SUSTech and HKU team EngineAI >RMB 10 billion USD 200 million Series B (with Luxshare Precision as strategic investor), Apr 2026 Full-size humanoid, focused on natural human-like gait DEEP Robotics ~RMB 7.89 billion Completed a Series C of over RMB 500 million; IPO coaching filing accepted Quadrupeds and humanoids, industrial applications; direct competitor to Unitree in quadrupeds and industrial scenarios Fourier Intelligence ~RMB 7.89 billion Completed shareholding reform; GR-series humanoids Began in rehabilitation medicine, extending into general humanoids Agility Robotics ~RMB 14.6 billion (SPAC listing at ~USD 2.5 billion) Digit has moved over 100,000 totes at GXO; agreement with Toyota Canada The US company with the deepest deployment mileage Boston Dynamics Under Hyundai Motor Electric Atlas delivering initial units to Hyundai; partnering with Google DeepMind on the brain A cautionary tale of top-tier technology and commercialisation failure 1X Technologies — Neo Gamma in retail pilots in Norway Soft-bodied robots for the home Apptronik ~RMB 38 billion Apollo in pilots at Mercedes and GXO, about 50 units Enterprise humanoid; the US "number three"

One brutal fact is buried in this table: in the ten-billion-valuation club, the overwhelming majority of companies are less than three years old (AgiBot, Galbot, Galaxea, AI² Robotics, RobotEra and X Square Robot were all founded in 2023, and Spirit AI in 2024). This is an industry that is startlingly young — and it has not yet been tested by a single genuine cycle. Boston Dynamics' warning is worth remembering for every Chinese company: it has worked on quadrupeds and humanoids for 34 years, with world-leading technology, yet never cracked commercialisation — acquired at a low price by Hyundai Motor Group for USD 1.5 billion in 2020, with its valuation nearly halved over a decade. Technological leadership has never equalled commercial success.

17. Cross-Sectional Comparison Table

Dimension Unitree AgiBot UBTech Galbot Figure AI Tesla Optimus Founded 2016 2023 2012 2023 2022 Programme from 2019; team earlier Technology core Body plus motion control (full-stack in-house components) Body, platform, Huawei-lineage AI Full-stack body plus industry solutions Embodied large model VLA brain plus body FSD neural networks plus manufacturing Flagship products G1/H2/R1; Go2/B2/A2 Expedition A3 / Lingxi X2 / Genie G2 Walker S2 / C1 Galbot S1/G1 Figure 03 Optimus Gen 3 2025 humanoid shipments 5,500+ (world No. 1) 5,100+ Several thousand ~1,200 Several hundred No external shipments H1 2026 revenue RMB 1.152 bn (+48.5%) Full-year 2025 over RMB 1 bn RMB 1.269 bn (+104%) — Not disclosed None Profitability Profitable (2025 net profit RMB 288 m) Not profitable Loss of RMB 790 m (2025) Not profitable Not profitable Not profitable Valuation / market cap Offer RMB 60.9 bn → peak RMB 444.9 bn → below RMB 200 bn ~RMB 15–18 bn (unlisted) ~HKD 42–46 bn ~RMB 21–22.5 bn ~USD 39 bn Consolidated into Tesla Core bet Hardware cost reduction and scale Platform, ecosystem and capital Industry orders and full stack Model generalisation Software entry point Manufacturing scale Greatest strength Cost, shipments, profitability Capital operations, Huawei lineage B2B orders, first to list Brains, national backing Real customers, brains Manufacturing, capital, captive scenarios Greatest weakness Weak "brains", high research-customer concentration, overstretched valuation Unclear path to profit Prolonged losses Small shipments Small production scale No external customers, lagging AI

18. The User Perspective: What Are People Actually Buying Unitree For?

The hardest part of cross-sectional comparison to get right is the voice of users — how they actually use the products, what they praise, and what they complain about. 18.1 Consumer robot dogs: a crowd magnet, and a great many idle owners An American journalist bought a Unitree Go2 for USD 4,000 in June 2026 and wrote an account. The opening is idyllic: he walked two miles to work each day with the robot dog beside him, from the Mount Pleasant neighbourhood of Washington DC to an office near the White House. He was photographed constantly; children queued to high-five the robot; real dogs barked at it. "This might be the world's most important robotics company," he wrote. The ending is stark. On the return journey, with the temperature at 30°C and the route mostly uphill, the robot's internal temperature hit 84°C (183°F) and its battery fell to 5 per cent — and within sight of his front door it suddenly toppled backwards, legs in the air. When asked what the machine was good for, his answer was: "the honest answer is not much." This is not an isolated case. Domestic user feedback is similarly telling: • Endurance is a structural weakness: early RMB-10,000-class robot dogs generally ran for only one to two hours, and many users complained that they "collapse after an hour or so of play"; only the 2026 models pushed endurance beyond four hours. • High idle rates: a 2025 GGV Robotics industry survey found that average monthly active use among owners of consumer quadrupeds in China was below 15 per cent — most robot dogs spend the overwhelming majority of their time gathering dust. • Expensive repairs: the standard whole-machine warranty is only one year. Out of warranty, third-party routine repairs cost RMB 1,000–1,500 (10–15 per cent of the purchase price); replacing a foot-pad rubber sleeve costs RMB 400. Core parts are entirely proprietary, with no third-party equivalents. • Interaction falls short: a living room is not a laboratory — between the television, children and kitchen noise, the robot dog simply cannot make out commands; in low light its camera perception fails; on complex terrain its follow function stalls. Yet another group of users adores it: event planners, mall operators, educational institutions and content creators. One user who rented a G1 wrote that at the entrance to an exhibition stand the robot produced "a near-100 per cent head-turn rate", that "children went from frightened to queuing to high-five it", and that "dwell time at the event rose by at least 30 per cent". A half-day rental costs a few thousand renminbi — "compared with several thousand to tens of thousands for a human performer for a full day, the value is actually good." The "same model as the Spring Festival Gala" tag is itself the most powerful traffic hook. 18.2 The mismatch between user perception and official positioning The mismatch is important: • Official positioning: Unitree describes itself as a "high-performance general-purpose robotics company", emphasising industrial inspection, emergency rescue and production-line work. • Actual usage: consumer — performance, spectacle, traffic, rental; research — running algorithms, publishing papers; industrial — pilot projects, demonstrations, a small volume of genuine work. • The true 2025 revenue mix: about 70 per cent from research and education, only about 9 per cent from industrial applications. In the public mind, therefore, "a Unitree robot" is closer to "a very cool performance machine" than to "an industrial device that does work". That perceptual mismatch is the deeper reason its valuation has reverted from "the master of the future factory" to "the star performer of the present stage". 18.3 How competitors' users see them • AgiBot: strong fan economy (Zhihui Jun's 2.69 million Bilibili followers) and high public trust; scepticism centres on "a high share of performance orders" and "the quality of revenue". The eight-hour factory livestream showing costs below human labour partially reversed that perception. • UBTech: relatively solid B2B reputation (BMW, Airbus, BYD), but the capital market is dissatisfied with "persistent losses and slow delivery on the story". • Figure: a rare Western example with genuinely paying customers; BMW's expansion is the strongest endorsement. Western media nonetheless question what proportion of the 200-hour livestream represented real productive work. • Tesla: Musk himself concedes that Optimus is "mainly learning and collecting data, rather than performing productive tasks" — when even the founder will not oversell it, user expectations are more rational in consequence. • Boston Dynamics: revered by users and media as "the god of technology", but dogged for over a decade by the label "admired but not bought".

19. Niche Analysis: Where Unitree Sits in the Sector Map

Picture the embodied-intelligence sector as a three-layer stack: ┌─────────────────────────────────────────────┐ Upper │ "Brains": embodied large models / VLA / │ │ world models │ │ e.g. Figure (Helix), Galbot, Galaxea, │ │ Spirit AI, X Square Robot, AI² Robotics │ ├─────────────────────────────────────────────┤ Middle│ "Body + platform": integrated machines plus │ │ operating systems and ecosystems │ │ e.g. AgiBot, UBTech, Tesla, LimX Dynamics, │ │ RobotEra, EngineAI │ ├─────────────────────────────────────────────┤ Lower │ "Hardware + cost": body manufacturing / │ │ joints / dexterous hands / sensors │ │ e.g. Unitree, DEEP Robotics, Fourier, │ │ Linker Hand, PaXini, Dobot │ └─────────────────────────────────────────────┘

Unitree's niche is "king of the lower layer". It is globally first in body manufacturing and cost control, first in shipments, and the most profitable company in the industry. The gap it fills is critically important — it has proven that humanoid robots can be cheap, can be mass-produced and can be profitable, establishing industry-wide confidence that "the hardware is viable". Yet its position is also delicate: • Downward, it holds an overwhelming cost advantage and supply-chain integration capability that nobody can shake in the short term; • Upward, it has almost no presence in the "brains" layer — its embodied large models are only just beginning (a dual-track WMA/VLA strategy, WVLA 2.0, the UnifoLM series), and its self-developed general embodied model has yet to be applied at scale in production models, being tested only in limited pilot scenarios such as its own factory; • Laterally, it fights hand-to-hand with AgiBot on shipments, overlaps with UBTech in industrial scenarios, and is misaligned with Galbot on technological narrative. The single most dangerous point is this: if companies in the "brains" layer (Figure, Galbot, Tesla) break through to general-purpose capability first, they can integrate the body from above — just as, in the smartphone era, the companies that owned the chip and the operating system ultimately defined the device. At that point, Unitree's prized "hardware cost moat" would become the fate of a contract manufacturer. But equally, if "brains" prove slow to break through, then whoever controls the lowest-cost, highest-volume body becomes the industry's infrastructure. That is precisely what Unitree is betting on.

20. Trend Assessment: Where Unitree Goes from Here

20.1 Opportunities

1. World-class hardware cost reduction: a core-component self-development rate above 90 per cent and high gross margins (63.2 per cent on humanoids) constitute a genuine, tangible moat;

2. Position as the shipment benchmark: first globally in both humanoids and quadrupeds, the industry's most concrete proof of "mass production demonstrated";

3. The scarcity of profitability: one of very few profitable integrated manufacturers, with greater resilience than the "cash-burning" camp (UBTech, AgiBot);

4. Determination on "brains": RMB 2.022 billion of IPO proceeds allocated to model R&D — nearly eight times the preceding three years' R&D combined — a correct direction;

5. Policy and industrial tailwinds: China accounts for 97 per cent of global humanoid shipments; the 15th Five-Year Plan explicitly designates embodied intelligence as a new growth engine among future industries; and in February 2026 China issued its first national-level Humanoid Robot and Embodied Intelligence Standards Framework (2026 Edition).

20.2 Risks

1. Continued reversion after an overstretched valuation: even after halving, on 2025 net profit of RMB 288 million the P/E remains in the hundreds; the bubble may not deflate in a single movement;

2. The lock-up flood: on 19 August 2027, approximately 229 million shares — 56.6 per cent of total share capital — will unlock — the single largest variable hanging over every existing shareholder;

3. The "brains" gap: an R&D expense ratio of 8.53 per cent (far below UBTech's 25.4 per cent), invention patents below 8 per cent of the total, and a self-developed large model not yet in volume production;

4. A fragile revenue structure: 70 per cent from research and education, only 9 per cent from industry — research procurement depends heavily on government and university budgets, and its sustainability is questionable;

5. A tightening US market: the 1260H list plus the FCC Covered List mean future models cannot enter the US; the US share of revenue has already fallen from 19.54 per cent to 13.30 per cent;

6. A step down in growth: from +335 per cent in 2025 to +48.54 per cent in H1 2026, with adjusted net profit already down 19.34 per cent year on year — the "high-growth" narrative is fading;

7. Industry consolidation: against more than 200 embodied-intelligence companies and over 100 humanoid robot companies in China, rapid capital concentration among leaders will squeeze the second tier — and Unitree itself may be caught up in a "dimensional strike" from Tesla or Figure.

20.3 A one-sentence assessment Unitree's greatest certainty lies in "hardware"; its greatest uncertainty lies in "brains". It is today a company ranked first in the world in "body" and in the second tier in "brain". Whether it can defend a valuation in the hundreds of billions depends on whether, within the next two to three years, it can convert its advantage in "body" into "commercial value that actually works" — which is precisely the shared bottleneck of the entire industry that Unitree itself acknowledges.

Part Three · Longitudinal–Cross-Sectional Convergence: What Does RMB 200 Billion Actually Mean?

The preceding two parts did two things: longitudinally, they traced Unitree's decade from 1990 to 11 September 2026; cross-sectionally, they placed it at the same table as AgiBot, UBTech, Galbot, Figure and Tesla. It is now time to weave the two threads together.

21. The Destiny of the Longitudinal, the Mirror of the Cross-Sectional

The longitudinal tells us that Unitree's rise is, at its core, the triumph of cost reduction. From a RMB 200 bipedal robot, to the RMB 20,000 XDog, to the RMB 20,000 Go1, to the RMB 9,997 Go2, to the RMB 99,000 G1, to the RMB 39,900 R1 and the RMB 26,900 dual-arm humanoid — over ten years Wang Xingxing has repeatedly done the same thing: turning what others regard as a luxury into a product the masses can afford. This is not a story of "technological innovation" but of "engineering, supply chain and cost". The cost of that story is that it welded the company's DNA firmly to "hardware". When the market asks "what can your robot do?", Unitree's answer was for a long time "it can dance, it can somersault, it can carry a person up a slope, it can perform". It pushed "body" to its limit without growing a "brain" in step. The cross-sectional tells us that the entire industry's valuation logic is shifting from "body" to "brains" and "customers". Look at Figure — with a tenth of Unitree's shipments, its valuation is more than ten times Unitree's pre-IPO valuation. On what basis? The Helix large model and a real paying customer, BMW. Look at AgiBot — three years old, revenue just past RMB 1 billion, yet valued at RMB 15–18 billion and heading for Hong Kong. On what basis? A "Huawei lineage" AI gene and capital imagination. Look at Galbot — a company with no stable revenue, valued at RMB 20 billion. On what basis? Its position at the top of the value chain as an "embodied large model" player. In other words, at this juncture in 2026, the capital market is willing to pay a premium for "brains" and "customers", and a discount for "body". Unitree is facing precisely that discounting process. It is not a company that has deteriorated; it is a company that has been "reclassified" — from "the first listed humanoid robot company (scarce plus imaginative)" to "a company that sells robots (manufacturing valuation)". RMB 444.9 billion was the price of imagination; RMB 192.9 billion is the price of reality. And the reality is: revenue of RMB 1.7 billion, net profit of RMB 288 million, growth decelerating, and 70 per cent of revenue from research. The longitudinal destiny is a "hardware origin"; the cross-sectional mirror is a "brains premium" — and the misalignment between the two is the true source of that RMB 240 billion evaporation.

22. Three Questions That Must Be Answered

Question one: Is Unitree a hardware company or an AI company? By revenue structure it is a hardware company (humanoids, quadrupeds and components account for 98 per cent). By self-narrative it wants to be an AI company (RMB 2.022 billion allocated to models). This uncertainty of identity is precisely the source of its violent valuation swings. If it is treated as a hardware company, it should trade on the general equipment manufacturing sector's 38.56-times P/E — implying a market capitalisation in the tens of billions. If it is treated as an AI platform company, it can trade on the "dream multiple" of Figure or Galbot — implying hundreds of billions or more. RMB 200 billion sits squarely between the two identities. Question two: Will Unitree become "the next Boston Dynamics"? The risk in that analogy is world-class technology and failed commercialisation. But there is one fundamental difference: Boston Dynamics never truly reduced costs or achieved scale shipments, whereas Unitree has done both. Boston Dynamics' Spot sold for USD 75,000 and still lost money; Unitree's robot dog sells for RMB 10,000 and turns a profit. On that measure, Unitree is closer to "the BYD or Tesla of robotics" than to Boston Dynamics. But another analogy is more dangerous: if "brains" remain unbroken, will Unitree become "the Foxconn of robotics" — manufacturing the finest bodies for other people's AI brains? That risk is more realistic than becoming Boston Dynamics. Question three: Below RMB 200 billion, is it value or a trap? That depends on which future you believe in: • If you believe the humanoid "ChatGPT moment" arrives within two to three years, then Unitree is now "a discounted entry ticket"; • If you believe the truth is Wang Xingxing's "at the latest five to ten years", then the current valuation still discounts a decade ahead; Note one detail: Wang Xingxing himself revised the "ChatGPT moment" from "one to two years" to "two to three years, up to five to ten years". A founder's guidance is always the most honest risk disclosure.

23. Integrated Judgement: My View

First, Unitree's decline is a "valuation reversion", not a "collapse in fundamentals". The company's products, shipments, profitability and supply-chain capability have not changed fundamentally. What changed is the market's discount rate and its classification of the company. The journey from RMB 444.9 billion to RMB 192.9 billion is, in essence, the market switching from "pricing the company on the long-term story of a trillion-renminbi market" to "pricing the company on current shipments and profits". That switch is healthy, and even necessary — because it forces the industry from "telling stories" back to "watching for delivery". Second, the real watershed is not 2026 but 2027. Two critical milestones will arrive at once: • 19 August 2027, when approximately 229 million shares (56.6 per cent of total share capital) unlock — a genuine supply-and-demand stress test; • 2027–2028, when "whether the embodied large model can land in real industrial scenarios" will receive a phase-one verdict — if by then Unitree's industrial revenue share is still below 10 per cent, the "hardware company" label will be permanently fixed. If Unitree can prove, before those two milestones, that "robots do real work and earn real money", it can reclaim an "AI company" valuation; if it cannot, a valuation in the hundreds of billions will revert slowly but firmly towards the tens of billions. Third, the industry is switching from "narrative-driven" to "delivery-driven", and Unitree stands precisely at the centre of that switch. The Humanoid Robot Industry Development Report released at the 2026 World Robot Conference stated that industry applications have "entered deep water at full scale". At the same conference, however, RUNTO noted that some 65 per cent of devices are still used for R&D, testing and exhibition demonstrations, with fewer than 35 per cent genuinely deployed in routine industrial operations. "Mass production" does not equal "mass commercial deployment" — this is the single sentence this industry most needs to internalise. And Unitree is precisely the company that has done "mass production" best. Its problem is that what it mass-produces is "robots that can perform", not "labour that can replace people". The gulf between those two is the gulf Unitree — and the entire industry — must cross over the coming five to ten years. Fourth, and most importantly: do not use short-term candlesticks to define the future of a hard-tech company, but do not use long-term stories to defend a price divorced from fundamentals. Unitree spent a decade proving that Chinese manufacturing can turn "impossible" into "cheap"; it now needs to spend the next decade proving that these cheap robots genuinely work and that someone genuinely pays for them. Until then, RMB 200 billion and RMB 100 million are both merely products of market sentiment, not value itself. Wang Xingxing said something in his TIME cover interview that remains the best footnote today: "Breakthroughs in hard-core technology take time."

The market priced that sentence at RMB 240 billion in eleven trading days.

Appendices

Appendix A: Unitree Key Timeline

Date Event 1990 Wang Xingxing born in Yuyao, Ningbo 2009 Enrolled at the School of Mechanical Engineering and Automation, Zhejiang Sci-Tech University 2015 Developed the electric-drive quadruped XDog during his master's; second prize in the Xingchuangshi competition (RMB 80,000) Jun 2016 Joined DJI as a motor control engineer 26 Aug 2016 Left DJI and founded Hangzhou Unitree Technology Co., Ltd. (angel investment of RMB 2 million from Yin Fangming at a post-money valuation of ~RMB 13.33 million) Sep 2017 Laikago, the first commercialised quadruped, launched 2019 Aliengo launched; Sequoia China led the Pre-A round (post-money RMB 150 million) Jan 2020 Educational quadruped A1 launched at CES Feb 2021 24 A1 robots appeared on CCTV's Spring Festival Gala for the Year of the Ox Jun 2021 Consumer quadruped Go1 launched Feb 2022 109 Go1 robots appeared at the Beijing Winter Olympics opening ceremony 12 Jul 2023 Consumer quadruped Go2 launched (from RMB 9,997) Aug 2023 First general-purpose humanoid H1 launched Nov 2023 Industrial quadruped B2 launched 13 May 2024 Humanoid G1 launched (from RMB 99,000) Feb / Sep 2024 Series B++ (valuation RMB 3.1 bn) / Series B+++ (valuation RMB 3.8 bn) 28 Jan 2025 16 H1 robots performed YangBot on CCTV's Spring Festival Gala Jun 2025 Series C of ~RMB 700 million at a post-money valuation of RMB 12.7 bn (Tencent, Alibaba, Ant, China Mobile and others) 25 Jul 2025 Humanoid R1 launched (from RMB 39,900) Aug 2025 Four gold medals at the inaugural World Humanoid Robot Games; quadruped A2 launched 20 Oct 2025 Full-size humanoid H2 launched Nov 2025 IPO coaching completed 20 Mar 2026 SSE accepts the STAR Market IPO application (raising RMB 4.202 bn) 1 Apr 2026 Selected for an on-site inspection of IPO applicants 16 Apr 2026 H1 breaks the human 1,500 m world record on a proportional basis at the marathon qualifier 1 Jun 2026 STAR Market IPO approved 8 Jun 2026 US Department of Defense adds Unitree to the 1260H list 2 Jul 2026 CSRC approves registration 28 Jul 2026 US FCC adds advanced robotic devices to its Covered List 6 Aug 2026 Offer price set at RMB 150.80 per share; offer market capitalisation RMB 60.9 bn 19 Aug 2026 Listed on the STAR Market; opened at RMB 1,100 (+629.44%), market capitalisation reaching RMB 444.9 bn 20 Aug 2026 Wang Xingxing deliberately cools sentiment at WRC; RMB 63.9 bn of market capitalisation evaporates in two days 2 Sep 2026 Share price falls below RMB 550 for the first time, halving from the peak 11 Sep 2026 Share price RMB 477.12, total market capitalisation RMB 192.978 bn — below RMB 200 billion

Appendix B: Principal Data Sources

• Unitree Technology prospectus and listing announcement (March–August 2026) • Cailian Press, Securities Times, Securities Daily, The Paper, Southern Metropolis Daily, 21st Century Business Herald, China Economic Net, Phoenix Finance, Sina Finance, Jiemian News, China News Service (2026) • 2026 Humanoid Robot Industry Development Report, 2026 World Robot Conference • Counterpoint Research, Omdia, IDC, GGV Robotics, RUNTO • Research notes from Huaxin Securities, GuoXin Securities, Guotai Haitong, CCB International, Nomura and Huatai Securities • Caijing, China Newsweek, Huanqiu.com, Legal Daily, China Economic News Network • UBTech (9880.HK) interim results announcement for the six months ended 30 June 2026 • RobotHub, The Robotic Life, RobotTesters, VentureAtlas, AI2Work, ValueAddVC (English-language sources)

Appendix C: Disclaimer

This report is a work of technology and business research. All data is drawn from public sources and is presented in good faith, though it may be subject to delay or error; figures relating to share price and market capitalisation should be taken as disclosed by the relevant exchange. The views expressed are analytical judgements and do not constitute investment advice of any kind. Any content labelled "estimated", "reported" or "unconfirmed" does not represent a finding of fact. Markets carry risk; invest with caut