Unitree: the world's cheapest robots, and the questions behind the hype
A neutral, evidence-first reading of China's Unitree Robotics — the volume leader in robot dogs and low-cost humanoids — assembled from English and Chinese primary sources and weighed question by question.
110 sources · 35% Chinese-languageAs of 31 May 202610 analysis sections
In under a decade Unitree went from a graduate-school robot dog to the world's best-selling maker of legged robots — and, on the back of a viral Spring Festival Gala, a 2026 candidate for a ~$6.2B Shanghai IPO.
The genuinely open question is not whether Unitree is impressive — its prices and shipment volumes are real and unmatched. It is whether radical hardware cost-leadership translates into a durable, software-defined, geopolitically-exposed business as the embodied-AI market either inflects or deflates. This study lays out both cases and weighs them: the disclosed demand mix and the IPO mark lean bearish, the hardware franchise leans bullish, and the software race is genuinely contested — the full weighing, with the tripwires that would flip each reading, closes the Forward View.
The decisive questions
Each links to the section that lays out the evidence on both sides.
Disclosed revenue from Unitree's IPO prospectus (¥ millions). The acceleration is both the bull case and, for those who see a hype peak, the bear case.
Weighed across the four questions: humanoid demand is still mostly research and display, not production (high confidence); the hardware cost moat is real but whether it wins a software-defined race is genuinely contested; the ~$6.2B IPO mark runs ahead of the disclosed fundamentals (medium confidence); and security-plus-geopolitics caps the Western upside without threatening the domestic core (medium confidence). The Forward View shows the work — the controlling evidence, the strongest counters, and the tripwires that would flip each reading.
How to read this
Ten sections, each built the same way: a neutral synthesis, a two-sided case-for / case-against ledger, dated quotes (with the original Chinese shown alongside any translation), interactive charts, and the sources used. Start with the question that interests you, or read in order from Company & Timeline.
🔍
Independent research artifact, not affiliated with or endorsed by Unitree Robotics. All quotes link to primary sources; private-round valuations are reported estimates and labeled as such; disclosed prospectus figures are marked. Where the research could not verify a claim, the page says so. See Methodology & Limits.
Section 01
Company & Timeline
From a graduate-school robot dog to a national champion and 2026 IPO candidate — in under a decade.
15 sources5 Chinese-languageAs of 31 May 2026
In nine years Unitree became the world's best-selling maker of robot dogs and the cheapest production humanoids, then a 2026 STAR Market IPO candidate — a rise propelled by genuine cost engineering and by viral national stagecraft [5][50][76]. How much of the story is each is the question this case study keeps returning to.
A nine-year arc
2016Wang Xingxing founds Unitree (Hangzhou Yushu Technology) on a ¥2M angel cheque [1][81].
2017Laikago research quadruped announced at ~$20–30K [4].
2021Go1 — a ~$2,700 consumer robot dog — and a Series A led by Shunwei [86].
2023Go2 launches from $1,600; the H1 humanoid debuts [28][33].
2024G1 humanoid unveiled at ~$16K; B2 round (~¥1B) closes [31][80].
Jan 202516 H1 robots dance at the CCTV Spring Festival Gala; inquiries surge [5][6].
Feb 2025Wang speaks at Xi Jinping's private-enterprise symposium [9].
Jun 2025Series C (~¥700M) closes at a ~¥12B valuation [79][87].
Mar 2026Files a 363-page STAR Market prospectus to raise ~¥4.2B [76].
Unitree Robotics (宇树科技; legal name Hangzhou Yushu Technology) was founded in August 2016 by Wang Xingxing (王兴兴), who had built the 'XDog' quadruped during a mechatronics master's and briefly worked at DJI before an angel cheque let him go full-time [1][2][81]. The product line walked steadily down the cost curve: the Laikago research dog (2017, ~$20–30K), the breakthrough consumer Go1 (2021), the Go2 (2023, from $1,600), and the industrial B2 — then a leap into humanoids with H1 (2023) and the $16K G1 (2024) [4][28][31].
The inflection was as much cultural as technical. On 28 January 2025, sixteen H1 robots danced a Yangge routine on China's CCTV Spring Festival Gala under director Zhang Yimou, and purchase inquiries surged [5][6]. Unitree robots then appeared at Beijing's humanoid half-marathon and a CCTV-broadcast kickboxing tournament, and Wang became the only post-90s founder to speak at Xi Jinping's February 2025 private-enterprise symposium — cementing Unitree as a state-blessed national champion and a member of Hangzhou's 'Six Little Dragons' [7][8][9][12][15].
Skeptics read the same events differently — as marketing. Unitree reportedly paid around ¥100 million to secure its Gala slot, and some Chinese commentators dismissed the performances as remote-controlled spectacle rather than autonomy [13]. The company is small for its fame — roughly 500 employees, hiring across every function as of 2025 — which both bulls and bears cite [1][10].
Both sides of the ledger
Both cases in full. This study's own weighing of them — leans, confidence and tripwires — is in the Forward View.
The case for
+Genuine category leadership: the world's best-selling robot-dog maker (~70% of global quadruped volume) and a top-2 humanoid shipper, all from a 2016 start [50][51].
+A credible low-cost engineering story dating back to a ¥200 first robot — Wang frames design, not scale, as the cost lever [3][11].
+State and capital endorsement: the Gala, the Xi symposium, 'Six Little Dragons' status, and a 2026 IPO on deck [5][9][12][76].
+Real commercial pull after the Gala — inquiries spanning performance, catering and factory uses, not just hobbyists [6].
The case against
−Much of the fame rests on stagecraft: a reported ~¥100M Gala spend and demos critics call remote-controlled spectacle [13].
−Public showpieces have stumbled — Unitree's G1 fell at the Beijing half-marathon start [14].
−It is small (~500 staff) and 'severely understaffed' by its own account — execution risk as it scales [1][10].
−The 'national champion' framing invites geopolitical scrutiny that a pure consumer-tech firm would avoid (covered under Risks) [95].
In their words
“If you think about it nonstop for 24 hours, I don't believe a problem can't be solved.”
original · zh“你24小时不停地去想，我不信问题解决不了”
Wang Xingxing (王兴兴) · Founder & CEO, Unitree · Feb 2025 · English is a translation from zh · source
“In the end, with just 200 RMB, I made a small bipedal robot.”
Wang Xingxing · recalling his first robot · — · source
Sources for this section
15 sources · en, zh · tiers shown. Full bibliography on the Sources page.
One of the most-hyped and least-proven markets in technology — with a decisive Chinese policy and supply-chain advantage.
12 sources3 Chinese-languageAs of 31 May 2026
The legged-robotics / embodied-AI market pairs enormous long-run forecasts (Goldman: $38B by 2035; Morgan Stanley: ~$5T by 2050) with thin, contested near-term demand — and a decisive Chinese policy and supply-chain advantage that Unitree is built to exploit [16][17][23].
📐
Forecasts are projections, not facts
Market-size figures below are forecasts from named houses (Goldman, Morgan Stanley, Interact Analysis) on different horizons. They disagree by orders of magnitude — shown as a range, not a single number.
China's structural edge in the value chain
Share of the global humanoid-robot component supply chain, per MERICS [23]. Unitree's cost advantage is partly an inherited national one.
Humanoid component supply chain (share of key companies)
China — 63%
Rest of world — 37%
The bull framing is a generational platform shift. Goldman Sachs raised its 2035 humanoid forecast sixfold to $38 billion; Morgan Stanley sketches a ~$5 trillion market by 2050 with a billion-plus units [16][17]. China is structurally advantaged: MERICS estimates it controls ~63% of the humanoid component supply chain and ~90% of permanent-magnet processing, and McKinsey notes Tesla's Optimus would cost roughly 3× more without Chinese suppliers [23][26]. Policy is explicit — MIIT's 2023 humanoid guidance and the 15th Five-Year Plan put 'embodied intelligence' alongside quantum and 6G [19][20].
The bear framing is that almost none of this revenue exists yet. Interact Analysis projected only ~40,000 humanoid units and ~$2B globally by 2032; Goldman's own November-2025 field research found suppliers building 100,000–1,000,000-unit capacity with no confirmed large-scale orders[18][22]. Chinese investors have openly called the sector a bubble with 'valuations that mortgage the future' [25].
Unitree's actual base is narrower and more real: the quadruped market, where China sold ~38,000 units in 2024 and Unitree leads. That segment is smaller and less glamorous than humanoids — but it ships, generates margin, and is where Unitree's dominance is least disputed [24][50].
Both sides of the ledger
Both cases in full. This study's own weighing of them — leans, confidence and tripwires — is in the Forward View.
The case for
+Forecasts are vast and rising — $38B by 2035 (Goldman) and a ~$5T 2050 vision (Morgan Stanley) [16][17].
+China's supply-chain and cost edge is real and quantified (~63% of components; Optimus 3× costlier without it) [23][26].
+A rare full-throated policy tailwind — national plans elevate embodied intelligence to strategic priority [19][20].
+Unitree's core quadruped market already ships at scale (~38K China units in 2024) and is growing >50% CAGR [24].
The case against
−Near-term demand is tiny and unproven — ~40K units / ~$2B by 2032 in a sober forecast [18].
−Capacity is being built ahead of orders; Goldman found no confirmed large-scale demand [22].
−Respected Chinese investors call it a bubble — froth, unclear scenarios, immature tech [25].
−The biggest forecasts are for humanoids, exactly where the real-world use case is least settled [18][108].
In their words
“The global market for humanoid robots could reach $38 billion by 2035 — more than a sixfold increase over our previous estimate of $6 billion.”
Goldman Sachs Research · industry forecast · 2024 · source
“No company has confirmed receiving large-scale orders or a clear production timeline.”
Goldman Sachs field research · on Chinese suppliers · Nov 2025 · source
Sources for this section
12 sources · en, zh · tiers shown. Full bibliography on the Sources page.
The lowest-priced legged robots in the field — with a persistent gap between demo polish and reliable autonomy.
15 sources4 Chinese-languageAs of 31 May 2026
Unitree builds the cheapest credible legged robots in the world — Go2 from $1,600, G1 from $13.5K — by designing >90% of components in-house. The recurring caveat, from reviewers and Unitree's own prospectus, is the gap between demo polish and reliable autonomous work[28][30][37].
The price chasm — flagship quadrupeds
List/reported prices, US$ thousands. Unitree's Go2 starts at $1,600; rivals sell at 25–125×. This single fact explains most of Unitree's volume lead [28][54].
Flagship quadruped list price (US$ thousands)
Unitree Go2
$2.8K
DEEP X30
$65K
BD Spot
$74.5K
Ghost V60
$150K
ANYbotics X
$200K
Industrial/defense models are contact-for-quote; figures are reported estimates where noted. Hover a bar for the basis.
The portfolio runs from the Go2 quadruped (from $1,600) and industrial B2 ($100K) to the G1 humanoid (US$13.5K, ~35 kg, 23 joints), the budget R1 ($5,900) and the 2026 H2 ($29,900) [28][34][30][32][39]. IEEE Spectrum called the Go2 'shockingly low cost,' and an independent motor teardown judged its in-house actuator 'thoughtful rather than cost-cutting' — Unitree designs >90% of its core parts, the root of both its price and its margin [29][35][37].
Capability claims are real but should be read carefully. The H1's 3.3 m/s speed mark is a category-specific, Unitree-claimed record, not an independently verified one; iFixit praised the Go2's repairability but flagged a fragile neck under falls [33][36]. The harder limit is software: analysts argue many viral humanoid clips are scripted, the budget R1 is explicitly a research platform — 'don't expect it to do your laundry' — and Chinese reporting notes the robots still lean on human teleoperation, with open-task AI latency over 5 seconds[32][38][42].
Crucially, Unitree's own IPO prospectus concedes it 'has not fully grasped' industrial customers' specific use cases — an unusually candid admission that the hardware has outrun proven applications [37].
Both sides of the ledger
Both cases in full. This study's own weighing of them — leans, confidence and tripwires — is in the Forward View.
The case for
+Unmatched price-performance: Go2 from $1,600, G1 from $13.5K — an order of magnitude below Western peers [28][30].
+Deep vertical integration (>90% in-house parts) validated by an independent teardown calling the engineering 'thoughtful' [35][37].
+A broad, fast-iterating lineup across quadrupeds and humanoids, down to a $5,900 R1 and ¥9,997 Go2 entry [32][40].
+Genuine, if narrow, capability records (e.g., H1 full-size-humanoid speed) [33].
The case against
−Headline feats are demo-grade: the speed record is self-claimed and category-specific; many humanoid clips look scripted [33][38].
−Durability questions — iFixit flagged neck fragility; budget models are explicitly not turnkey [36][32].
−The real bottleneck is AI/software generalization, where Unitree is not the leader [62].
−Unitree itself admits it hasn't fully grasped industrial customers' use cases [37].
In their words
“We have researched and developed almost every mechanical part.”
A profitable hardware seller at ~60% margin — but most humanoid demand is research and display, not production.
7 sources4 Chinese-languageAs of 31 May 2026
Unitree is a profitable hardware seller at ~60% gross margin (rare in this field) with low customer concentration — but ~74% of humanoid revenue comes from research/education buyers, and much commercial demand is rental-and-display rather than production work [43][45][47].
Where the revenue comes from
Revenue mix, first nine months of 2025 (% of total), from the prospectus [53]. Humanoids overtook quadrupeds for the first time.
Revenue mix, 9M-2025
Humanoid robots — 52%
Quadruped robots — 42%
Components & other — 6%
The demand-quality question
Within humanoid revenue, by buyer type (% of humanoid revenue) [45][49]. Research and education dominate; genuine industrial production is a sliver.
Humanoid revenue by buyer type, 9M-2025
Research / education — 74%
Commercial / consumer — 17%
Industrial production — 9%
A unit walk from the disclosed inputs (illustrative)
Simple division over prospectus figures gives a rough unit picture for the nine months to September 2025. Total revenue was ¥1.167B [74], of which humanoids were just over half (>51%) [53] — call it ~¥0.6B. At the disclosed humanoid ASP of ~¥168K [44], that implies on the order of ~3,500 humanoids sold in nine months, consistent in scale with the disputed 4,200–5,500 full-year shipment figures [51][52]. Against that, revenue from genuine industrial deployment was ~¥15.7M [49] — at the same ASP, on the order of ninety robots doing production work, about 1.3% of total revenue. Derived figures are illustrative; the inputs are cited and the method is division.
The model is refreshingly simple: design and build the robots, sell them, and keep margin high by making the expensive parts yourself. Main-business gross margin reached 60.13% in 2025 — external procurement is under 4% of production cost — and customer concentration is low, with the top-5 customers just ~10.6% of revenue [43][44][59]. That diversification and margin profile is genuinely unusual for a young robotics company.
The quality of demand is the open question. Within humanoid revenue, 73.6% comes from research and education buyers; only 9% is industrial production [45]. Per the prospectus, many non-academic buyers deploy robots 'for show' — promoters, exhibitions, tourist sites — and rental economics weakened sharply: G1 day-rates reportedly fell from ~¥10,000 to ¥1,000–3,000 through 2025 as the post-Gala rush faded [47][46].
So the bull and bear cases share a fact base: real margins and real sales, but demand skewed toward experimentation and spectacle rather than durable production deployments — the gap Unitree must close to justify its valuation.
Both sides of the ledger
Both cases in full. This study's own weighing of them — leans, confidence and tripwires — is in the Forward View.
The case for
+~60% gross margin on hardware, driven by in-house components — a genuinely strong unit-economic base [43][59].
+Profitable and cash-generative while many humanoid peers burn cash (see Financials) [75].
+A landmark ¥124M China Mobile order signals the start of institutional procurement [48].
The case against
−~74% of humanoid revenue is research/education; by one read of the filing only ~2.6% is genuine industrial deployment [45][49].
−Much commercial demand is 'for show' (promoters, exhibitions), not value-creating work [47].
−Rental rates collapsed post-Gala (≈¥10,000 → ¥1,000–3,000/day), lengthening buyer payback and signalling soft real demand [46].
−Falling humanoid ASPs (¥593K → ¥168K) mean volume must rise fast just to hold revenue [44].
In their words
“Non-academic consumers who buy these robots are mostly deploying them 'for show': as attractive promoters in retail settings, at tourist sites, in performances and exhibitions.”
“The investment could once be recovered within a month, but with prices dropping from 10,000 yuan to 1,000 yuan, the payback period has significantly lengthened despite unchanged costs.”
The Standard (Hong Kong) · on humanoid rental economics · Mar 2025 · source
Sources for this section
7 sources · zh, en · tiers shown. Full bibliography on the Sources page.
Dominant in quadrupeds, top-tier in humanoid volume — but trailing on AI software in a crowded, fast-scaling field.
9 sources2 Chinese-languageAs of 31 May 2026
Unitree dominates quadrupeds (~70% of global volume) and is a top-2 humanoid shipper on price and speed — but it trails on AI software, and a domestic rival (AgiBot) out-shipped it in 2025 humanoids per Omdia, in a field crowded with far better-funded players [50][51][66].
Five Forces — legged & humanoid robotics
Click a force for the rated pressure and its evidence. Rivalry and new entrants are the binding constraints; Unitree's in-house components keep supplier power low.
Porter's Five Forces — Legged & humanoid robotics
Legged & humanoid robotics
Competitive rivalry — High pressure. China alone hosts 150+ humanoid firms; AgiBot out-shipped Unitree in humanoids in 2025 (Omdia) and reached its 10,000th unit by March 2026. In quadrupeds Unitree leads (~70% share), but DEEP Robotics, Boston Dynamics, Ghost and ANYbotics all compete. Price wars are already visible.
Positioning — price vs. capability
Unit price (left = cheaper) against demonstrated capability / AI autonomy (up = higher). Hover a point for the basis. Unitree owns the low-price corner; Western leaders own the capability frontier [50][66].
Humanoid / legged-robot positioning
Hover a point to see the basis for its placement.
In quadrupeds the picture is lopsided in Unitree's favour: SemiAnalysis estimates ~70% global volume share, roughly 10× the next competitor, against premium rivals like Boston Dynamics' $74,500 Spot [50][54]. The whole market is essentially Unitree's price story made concrete.
Humanoids are more contested. Omdia (via SCMP) ranked AgiBot No.1 in 2025 (5,168 units, 38%) and Unitree No.2 (4,200, 32%); IDC reached a similar order, while Unitree calls the circulating figures 'false information' and reports 5,500+ deliveries and the top spot [51][52][57]. Either way, Chinese firms shipped ~90% of global humanoids [23], and Unitree alone out-shipped US rivals Figure and Tesla by an estimated ~36× [55]. Against Western peers Unitree's edge is price and volume; their edge is software and proof — Figure (valued $39B) ran a real BMW line, a depth of deployment Unitree hasn't matched [66][67].
The competitive risk is that volume leadership is fragile when capital and AI talent flood in: AgiBot hit its 10,000th humanoid by March 2026, EV makers like Xpeng are entering, and price wars are already visible [56][73].
Both sides of the ledger
Both cases in full. This study's own weighing of them — leans, confidence and tripwires — is in the Forward View.
The case for
+Commanding quadruped lead — ~70% global volume, ~10× the next rival, at a fraction of Spot's price [50][54].
+Top-tier humanoid volume — out-shipping Figure and Tesla combined by ~36× [55].
+Unitree contests the No.2 ranking, reporting 5,500+ humanoid deliveries and No.1 [52].
+Price is a durable wedge in a market where rivals sell at 5–25× the cost [54][66].
The case against
−A domestic rival out-shipped Unitree in humanoids in 2025 per Omdia, and is scaling faster [51][56].
−Western peers lead on AI software and proven deployment — Figure's BMW line, $39B backing — and Chinese analysts call Unitree 'strong in body, weak in brain' [66][67][58].
−The field is crowded and well-capitalised (Figure, Tesla, Xpeng, UBTech), pressuring price and share [69][73].
−Volume leadership built on low price is vulnerable to a price war it may not win on capital [56].
In their words
“Unitree controls an estimated 70% share of global quadruped sales by volume… approximately 10x the total shipments of the next competitor.”
“In 2025, Unitree's actual shipment volume of humanoid robots exceeded 5,500 units… quantity actually sold and delivered to end customers, not order volume.”
Unitree Robotics · disputing the Omdia ranking · Jan 2026 · source
Sources for this section
9 sources · en, zh · tiers shown. Full bibliography on the Sources page.
A deep cost moat in hardware — possibly mis-located if the lasting advantage in embodied AI is software.
7 sources3 Chinese-languageAs of 31 May 2026
Unitree's moat is cost via vertical integration — >90% in-house parts, ~60% margin, and unmatched prices. The bear case: it's a hardware moat in a software race, with R&D at just ~7.7% of revenue and AI compute imported from Nvidia/Intel [59][60][63].
Value chain — what Unitree owns vs. depends on
Where the moat is deep (hardware) and where it is thin (AI cognition, compute, deployment) [59][62][63].
Generalisable cognition — the contested gap; ~7.7% R&D, model not yet at scale.
dependent
High-end compute
Nvidia/Intel imported via intermediaries — sanctions-exposed.
dependent
Deployment / apps
Industrial use cases 'not fully grasped' per the filing.
🧠
The crux
Unitree's moat is deepest exactly where hardware is commoditising fastest, and thinnest where the durable advantage in embodied AI may form. The IPO's 85%-to-R&D plan is a bet to close that gap [65].
The stated and revealed strategy align unusually well: be the cost leader by owning the bill of materials. External procurement is under 4% of production cost, motors-to-LiDAR are designed in-house, and Wang explicitly credits design (not just scale) for low cost [59][37]. That protects margin, enables aggressive pricing, and is hard for integrators who buy parts to match. Wang's macro bet is that robotics is at the EV industry's pre-takeoff moment — a 'trillion-yuan battlefield' [64].
The vulnerability is that the decisive moat in embodied AI may be software — generalising across messy real environments — not actuators. TrendForce flags exactly this gap, and Unitree's ~7.73% R&D intensity is modest for an 'AI-native' robot maker [62][60]. Tellingly, the IPO earmarks 85% of the raise for R&D — ¥2.02B (48%) for embodied-AI models — an explicit attempt to buy its way from hardware maker to full-stack AI platform [65]. It also depends on imported high-end compute (Nvidia Jetson, Intel) routed through intermediaries — a concentrated, sanctions-exposed dependency the IPO filing underplays [63].
So the moat is real but possibly mis-located: deep where hardware is commoditising fastest, shallow where the lasting advantage may form.
Both sides of the ledger
Both cases in full. This study's own weighing of them — leans, confidence and tripwires — is in the Forward View.
The case for
+Vertical integration (>90% in-house parts; <4% external procurement) is a genuine, margin-protecting cost moat [59][37].
+Cost leadership lets Unitree set the market's price floor and out-iterate part-buying integrators [61].
+A coherent founder thesis — robotics at the EV-style inflection — guiding aggressive, focused execution [64].
+Manufacturing and supply-chain depth in China compounds the cost edge [23][26].
The case against
−It's a hardware moat in a software race — generalisable AI, where Unitree isn't the leader, may decide it [62].
−R&D is only ~7.7% of revenue — thin for a company selling an AI future [60].
−Reliance on imported Nvidia/Intel compute via intermediaries is a sanctions-exposed dependency [63].
−Cost advantages erode as Chinese rivals copy the in-house-component playbook [56].
In their words
“Robotics is where EVs were a decade ago, a trillion-yuan battlefield waiting to be claimed.”
The price-and-profit outlier — far cheaper and actually profitable, but valued below AI-first Western leaders.
8 sources1 Chinese-languageAs of 31 May 2026
Across peers, Unitree is the price-and-profit outlier — far cheaper and actually profitable — but valued well below AI-first Western leaders like Figure ($39B) and out-shipped in 2025 humanoids by AgiBot[66][51].
2025 humanoid shipments (Omdia)
Units shipped, per Omdia via SCMP [51]. Unitree disputes this, self-reporting 5,500+ deliveries and the No.1 spot [52][57] — shown as a note, not the bar.
2025 humanoid robot shipments (Omdia estimate)
AgiBot 智元
5,168 units
Unitree 宇树
4,200 units
UBTech 优必选
1,000 units
Benchmark table
Flagship humanoid price, valuation/market cap, and posture. Figures are reported estimates unless from filings; prices vary by configuration.
n/d = not disclosed. Prices are list/reported and configuration-dependent.
On price the gap is stark: Unitree's Go2 starts at $1,600 versus Boston Dynamics' $74,500 Spot; its G1 humanoid lists at $13.5K against Agility's Digit at a reported ~$250K[28][54][68]. On profitability Unitree is the rare one in the black — roughly ¥600M non-GAAP 2025 profit — while UBTech and most Western humanoid makers run deep losses [75][69].
On valuation and AI, though, the West leads: Figure is valued at $39B with a real BMW deployment; Tesla brings a manufacturing flywheel and a ~$20K price target that would directly challenge Unitree [66][67][72]. Among Chinese peers, AgiBot (Tencent-backed, ~¥15B valuation) out-shipped Unitree in humanoids in 2025 and is also IPO-bound, while DEEP Robotics presses on quadrupeds and UBTech is already public [70][71][69].
The comparison table below makes the trade-off explicit: Unitree optimises for unit cost and volume; peers optimise for capability, AI, or capital. Which axis wins the decade is the unresolved question.
Both sides of the ledger
Both cases in full. This study's own weighing of them — leans, confidence and tripwires — is in the Forward View.
The case for
+Cheapest in class by far — Go2 at ~2% of Spot's price; G1 at ~5% of Digit's [28][54][68].
+Actually profitable (~¥600M non-GAAP 2025) where UBTech and Western peers lose money [75][69].
+Highest legged-robot volume of any peer, East or West [50][55].
+A valuation (~$6.2B target) that looks modest next to Figure's $39B for far more shipments [77][66].
The case against
−Valued far below — and behind on AI versus — Figure ($39B) and Tesla's flywheel [66][72].
−Out-shipped in humanoids by Tencent-backed AgiBot in 2025 [51][70].
−Peers show deeper real deployments (Figure at BMW; UBTech at BYD/Foxconn) [67][69].
Genuinely profitable and fast-growing — a rarity in humanoids — heading into a ~$6.2B Shanghai listing.
17 sources11 Chinese-languageAs of 31 May 2026
Unitree is profitable and fast-growing — revenue ¥392M (2024) → ¥1.708B (2025), with profit — heading into a ~¥42B ($6.2B) STAR Market IPO. The cautions: the mark tripled from the private round, headline profit is non-GAAP, and Q1-2026 profit already fell ~53% [74][77][83].
🧾
Disclosed vs. estimated
Revenue and profit below are from Unitree's audited STAR Market prospectus. Pre-IPO round valuations are reported estimates; the ~¥42B IPO figure is a target, not a closed mark.
Revenue trajectory (disclosed)
¥ millions, from the prospectus [74][75]. Revenue rose ~14× from 2022 to 2025.
Revenue, 2022–2025 (¥ millions, audited)
Valuation trajectory (estimated / target)
¥ billions. The IPO target is roughly 3× the mid-2025 private round — a steep step-up [79][77].
Post-money valuation / IPO target (¥ billions)
⚠️
Read the profit line carefully
The widely-quoted ¥600M 2025 profit is non-GAAP; GAAP net profit was ~¥288M, the gap largely stock-based comp. And Q1-2026 adjusted profit fell ~53% even as revenue grew 68% [75][83].
What the ~¥42B mark asks you to believe (illustrative)
The last closed private mark was the ~¥12B Series C of June 2025 [79]; the IPO targets ~¥42B (~$6.2B) [77] — a ~3.5× step-up in about eleven months. Divided over the disclosed full-year 2025 results (revenue ¥1.708B; GAAP net profit ~¥288M; non-GAAP >¥600M [75]), the target implies roughly 25× trailing sales and ~146× trailing GAAP earnings (~70× on the non-GAAP line) — simple division over prospectus inputs, stated here as the bar rather than a judgment. For the sales multiple to compress to ~10× at a flat mark, revenue must reach ~¥4.2B — about 2.5× 2025 — with the ~60% margin holding [89]; at TrendForce's projected ~94% 2026 industry pace that takes between one and two years of keeping share [106]. That is what both sides are measured against: a buyer of the mark must believe the +335% growth persists and that the Q1-2026 profit fall (~−53% [83]) is investment rather than erosion; a skeptic must explain why the field's only profitable pure-play should be worth a fraction of AI-first peers marked far higher pre-profit [66].
Disclosed figures are strong and, unusually, audited. Revenue rose ¥123M → ¥159M → ¥392M → ¥1.708B across 2022–2025 (+335% in 2025); the company turned its first annual profit in 2024 (¥94.5M) and reported a 2025 non-GAAP profit above ¥600M[74][75]. It filed a 363-page prospectus on 20 March 2026 to raise ~¥4.2B, with the listing-committee hearing set for 1 June 2026 — a fast 73-day track at a target valuation near ¥42B ($6.2B)[76][77][88].
Three caveats temper the story. First, valuation: the IPO target (~¥42B) is roughly 3× the ~¥12B Series-C mark from mid-2025 [79][77]. Second, profit quality: the headline ¥600M is non-GAAP — GAAP net profit was ¥288M, the difference largely stock-based comp [75]. Third, momentum: Q1-2026 revenue grew 68% but adjusted profit fell ~53% as R&D and competition bit and the hype cooled [83]. The company also receives meaningful state support (~¥76M tax breaks; ~¥32M grants) [82].
The funding history — angel ¥2M (2016) to a China-Mobile/Tencent/Alibaba-backed Series C — shows steadily escalating, blue-chip conviction; the IPO will test whether public markets share it [81][79].
Both sides of the ledger
Both cases in full. This study's own weighing of them — leans, confidence and tripwires — is in the Forward View.
The case for
+Audited, fast growth with real profit — ¥1.708B 2025 revenue (+335%), profitable since 2024 [74][75].
+Strong ~60% gross margins (60.13% in 2025, up ~16 pts since 2023) and low customer concentration underpin the numbers [43][89][44].
+Blue-chip backers (China Mobile, Tencent, Alibaba, Ant, Geely) and a fast-tracked IPO [79][77].
+Rare profitability in a field where most humanoid makers burn cash [69][75].
The case against
−IPO target (~¥42B) is ~3× the mid-2025 private mark — a steep step-up [79][77].
−Headline ¥600M profit is non-GAAP; GAAP profit was less than half that [75].
−Q1-2026 adjusted profit fell ~53% — early evidence margins are under pressure [83].
−Reliance on government grants/tax breaks flatters the picture and signals policy dependence [82].
In their words
“From 2022 to 9M-2025, revenue was ¥123M, ¥159M, ¥392M and ¥1.167B; net profit was −¥22.1M, −¥11.1M, ¥94.5M and ¥105M.”
original · zh“2022年-2025年前三季度，营收分别为1.23亿元、1.59亿元、3.92亿元和11.67亿元；净利润分别为-2210.05万元、-1114.51万元、9450.18万元和1.05亿元”
证券时报 (Securities Times) · from the prospectus · Mar 2026 · English is a translation from zh · source
“Adjusted net profit plummeting over 52 percent, attributed partly to a cooling of the broader humanoid robotics hype and increasingly fierce competition.”
An unusually broad risk surface for a hardware company — cyber, military/dual-use, geopolitics, and demand substance — with Unitree's responses shown alongside.
15 sources4 Chinese-languageAs of 31 May 2026
Unitree's risks are unusually broad: documented cyber-vulnerabilities (a Go1 backdoor; the wormable 'UniPwn'), well-evidenced military/dual-use exposure drawing US sanction calls, and the structural risk that current demand is more spectacle than substance[91][93][95].
🛡️
Attribution, not accusation
The critical claims here are attributed to named researchers, lawmakers and outlets, and paired with Unitree's own responses where they exist. Weighed: the security findings are technically documented and largely acknowledged by Unitree's own fixes (high confidence), while the military-tie allegations are evidenced but contested in scope — serious enough to move US policy, not adjudicated as company intent.
The risk surface, in brief
Cybersecurity: a documented Go1 "backdoor" (CVE-2025-2894) and a wormable, root-level "UniPwn" exploit across Go2/B2/G1/H1; Unitree shut the Go1 service and says most fixes are done [91][93][92][94].
Military / dual-use: a rifle-equipped Go2 in PLA propaganda, B2 robots in a 2024 China–Cambodia drill, sales to Entity-List universities, and a Kharon-documented partner network tied to a PLA "Robot Wolf" combat platform that Unitree declined to address; Unitree says any weaponisation is third-party and cites a 2022 pledge [96][98][97].
US policy: all 24 House Select Committee members urged listing Unitree; 2026 bills would curb federal use of Chinese robots — yet US buyers persist for lack of alternatives [95][101][102].
Demand substance: a marathon fall, an H1 flailing incident, and prominent VCs (Zhu Xiaohu) exiting and calling customers "imaginary" [14][99][104][105].
Security is the most concrete. Researchers disclosed CVE-2025-2894, an undocumented 'backdoor' tunnel in the Go1 giving remote access to ~1,919 devices; months later 'UniPwn' showed a wormable, root-level exploit across Go2, B2, G1 and H1 [91][93]. Unitree's response was mixed — it shut the Go1 service and said it had 'completed the majority of fixes,' but characterised the original feature as common and was criticised for slow engagement [92][94].
Dual-use is the most politically charged. A rifle-equipped Go2 appeared in PLA propaganda and B2 robots featured in a 2024 China–Cambodia drill; Kharon documents sales to Entity-List universities and a partner tied to a PLA combat platform [96][98]. All 24 members of the US House Select Committee urged listing Unitree on military/Entity/FCC lists, and proposed 2026 legislation would curb federal use of Chinese robots [95][101]. Unitree's rebuttal: it makes civilian products, disavows weaponisation (a 2022 pledge), and notes its filing omits military uses — even as it keeps selling into the US where few alternatives exist [97][102].
The quieter risks are commercial: a safety incident where an H1 flailed near handlers, the public G1 marathon fall, and prominent investors warning the demand is 'somersaults' without commercialization [99][14][100].
Both sides of the ledger
Both cases in full. This study's own weighing of them — leans, confidence and tripwires — is in the Forward View.
The case for
+Unitree acknowledged and patched the Go1 issue and says most UniPwn fixes are done — it is responding [92][94].
+It publicly disavows weaponisation (2022 pledge), says its products are civilian, and its filing omits military use [97].
+US demand persists despite the politics because credible non-Chinese options are scarce [102].
+Low customer concentration and diversified products cushion any single shock [44].
The case against
−Serious, repeated security flaws — a backdoor and a wormable root exploit across the whole fleet [91][93].
−Well-documented military/dual-use links drawing bipartisan US calls for sanctions/listing [95][96][98].
−Pending US restrictions could close a growth market and signal wider de-risking [101].
−Safety and substance doubts — an H1 flailing incident, a marathon fall, and prominent VCs (Zhu Xiaohu) exiting the sector, calling its customers 'imaginary' and warning of a 'valley of death' [99][14][100][104][105].
In their words
“An infected robot can simply scan for other Unitree robots in BLE range and automatically compromise them, creating a robot botnet that spreads without user intervention.”
IEEE Spectrum · on the UniPwn exploit · Sep 2025 · source
“Who are their potential clients? Who would spend tens of thousands of yuan to do these tasks?”
Zhu Xiaohu (朱啸虎) · GSR Ventures, on humanoid demand · Oct 2025 · source
Sources for this section
15 sources · en, zh · tiers shown. Full bibliography on the Sources page.
Three scenarios, then the weighing: where the evidence leans on each decisive question, the confidence behind it, and the tripwires that would flip it.
5 sources2 Chinese-languageAs of 31 May 2026
Three questions decide Unitree's decade: does humanoid demand become real, does its cost moat survive an AI-defined race, and does geopolitics wall off markets? The scenarios below are framed for you to weigh — not a prediction this study endorses [106][108][101].
Three ways the next few years could run
Bull case
Cost leader of a real platform shift
IPO capital funds AI + capacity, humanoid demand inflects (China output +~94% in 2026), and Unitree's price-and-volume lead compounds — Wang's EV analogy plays out [106][64][110].
Watch: Paid, repeat industrial deployments; a closing AI-software gap.
Base case
Profitable hardware leader, rich valuation
Quadrupeds keep paying the bills; humanoids grow steadily but slowly. Unitree stays profitable and #1-or-#2 by volume, but the mark runs ahead of near-term fundamentals [107][83].
Watch: Margin trend; AgiBot and EV-maker entrants; rental economics.
Bear case
Hype deflates, market walls off
Demand stays display/research, a price war compresses margins, and US restrictions plus security stigma cap the addressable market — a "valley of death" before any recovery [108][101][104].
On whether humanoid demand is real or mostly spectacle: the evidence leans spectacle-first — real production demand has not arrived yet (high confidence). The controlling evidence is the prospectus's own buyer mix — 73.6% of 9M-2025 humanoid revenue from research/education and only ~¥15.7M (~2.6%) from genuine industrial deployment [45][49] — and Goldman's field finding that Chinese suppliers are building 100,000–1,000,000-unit annual capacity with no confirmed large-scale orders [22], which outweighs the +480% industry shipment surge [51] because disclosed revenue composition measures what buyers actually pay for, while shipment counts include robots bought for lobbies and stages [108]. The strongest surviving counter-argument: state procurement has begun — China Mobile's ¥124M order, split between AgiBot and Unitree [48] — and research labs are the natural first buyers of any new computing platform. What would flip this reading: industrial-production share of humanoid revenue above ~25% in Unitree's first post-listing annual report (FY2026, due spring 2027), or humanoid rental rates recovering above ~¥5,000/day from the ¥1,000–3,000 trough [46]. Pre-mortem: if this looks wrong in two years, the most likely reason is that research buyers seeded a real industrial market faster than the composition data suggested — or, on the other side, that even the research wave was one-off, gala-driven stockpiling [6].
On whether a cost moat wins a software race: the evidence is genuinely contested. The hardware case is proven: ~60% gross margin on >90% in-house components [89] and ~70% of global quadruped volume [50], in a supply chain where McKinsey estimates a humanoid BOM roughly triples without Chinese suppliers [26]. But the software gap is conceded by Unitree itself — the prospectus admits its general embodied model is not deployed at scale [103], R&D ran only ~7.7% of revenue [60], and earmarking 85% of the raise for R&D is the company pricing its own deficit [65]. What deadlocks the question is that nobody has shown embodied-AI software is winner-take-most: if capable models commoditize, the cheapest body wins; if they don't, Figure's BMW deployment is the shape of the future [67]. What would flip this reading: Unitree shipping its own embodied model at scale in products by the FY2027 report, or an AI-first rival converting software into paid industrial deployments at a pace Unitree visibly cannot match. Pre-mortem: the bull-side miss would be underestimating what a ¥2.02B dedicated AI program buys [65]; the bear-side miss, overestimating how long ~60% hardware margins survive a price war already visible in Q1-2026 [83].
On whether the ~$6.2B IPO mark is earned or frothy: the evidence leans frothy relative to disclosed fundamentals (medium confidence). The controlling evidence is the ~¥42B target itself — ~3.5× the ¥12B Series C struck eleven months earlier [77][79] and roughly 25× trailing sales / ~146× trailing GAAP earnings on 2025 numbers [75] — set against Q1-2026 adjusted profit falling ~53% [83], which outweighs the genuine profitability and +335% growth because the step-up assumes the growth persists while the newest disclosed quarter points the other way. The strongest surviving counter-argument: Unitree is the field's only profitable pure-play, and AI-first peers carry far richer marks pre-profit — Figure at $39B [66] — so relative to its sector Unitree is the conservative bet. What would flip this reading: FY2026 revenue at or above ~¥3.4B (2× 2025) with gross margin holding near 60% at the spring-2027 annual report [89], or the listed stock sustaining the mark through its first lock-up expiry. Pre-mortem: if this looks wrong in two years, the most likely reason is that 2026's projected ~94% industry output surge [106] flowed disproportionately to the volume leader — or, on the other side, that a momentum listing met the "valley of death" investors like Zhu Xiaohu predicted [104].
On whether security and geopolitics cap the upside: the evidence leans toward a real but partial cap — Western upside narrows while the domestic core is untouched (medium confidence). The controlling evidence is the unanimous House Select Committee push to list Unitree [95] and 2026 bills restricting federal use of Chinese robots [101], set against the fact that the demand engine is at home: China runs 54% of global industrial-robot installations [27] and the 15th Five-Year Plan puts embodied AI on the strategic tier [20], which outweighs the stigma because US buyers keep purchasing anyway for lack of alternatives [102]. The strongest surviving counter-argument: the wormable UniPwn exploit across four product lines [93] gives any Western procurement ban a concrete technical basis, not just politics. What would flip this reading: an actual BIS Entity-List addition or passage of the American Security Robotics Act / FY27 NDAA restrictions [101] — or, the other way, a clean independent security audit plus continued US revenue growth. Pre-mortem: the bull-side miss would be assuming Washington stays slower than the market; the bear-side miss, treating Western sales as load-bearing when the business is funded from China.
Net: a real company with a real cost franchise, priced for an inflection its own disclosed demand mix does not yet show. The base case above is the modal reading; the bull case needs the demand answer to flip; the bear case needs the valuation and geopolitics answers to land together.
Bull path: the IPO funds an AI and capacity push, humanoid demand inflects (TrendForce sees China output up ~94% in 2026), and Unitree's cost leadership turns into durable share as the market scales — the EV analogy Wang invokes, alongside his claim that 2026 humanoids will run a sub-10-second 100m [106][64][110]. Base path: quadrupeds keep paying the bills and humanoids grow steadily but slowly, leaving Unitree a profitable hardware leader whose valuation runs ahead of near-term fundamentals [107][83].
Bear path: the hype deflates. A sober 2035 forecast still sees only ~700k units / ~$15B; new buyers are state firms parking robots in lobbies; margins compress as a price war with AgiBot and others intensifies; and US restrictions plus security stigma cap the addressable market [107][108][101]. A Jan-2026 regulatory wobble around its IPO 'green channel' hinted at official wariness of froth [109].
What to watch: humanoid units that do paid, repeat industrial work (not displays); R&D intensity and any closing of the AI-software gap; the pace of AgiBot and EV-maker entrants; and the trajectory of US/EU policy. The evidence today genuinely supports more than one ending.
Both sides of the ledger
Both cases in full. This study's own weighing of them — leans, confidence and tripwires — is in the Forward View.
The case for
+If demand inflects, Unitree's cost + volume lead compounds as the market scales [106][64].
+IPO capital can fund the AI and capacity investment the bear case says it lacks [76].
+Quadrupeds provide a profitable base to fund the humanoid bet through a slow patch [50][75].
+China's policy and supply chain keep a structural tailwind behind it [23][20].
The case against
−Sober forecasts see humanoids small for years (~$15B by 2035) — valuation may be early [107].
−Current demand skews to display/research, not the production work that justifies scale [108].
−A price war with AgiBot and EV entrants could erode the cost moat and margins [56][83].
−US restrictions and security stigma could cap the addressable market [101][95].
In their words
“The main buyers used to be academic research labs. Now we have a new customer profile: state-owned enterprises putting them in lobbies for display.”
KrASIA · on the demand reality check · Dec 2025 · source
“China's humanoid robot output to surge 94% in 2026.”
What this case study is, how it was researched, what is disclosed vs. estimated, and where it may be wrong.
110 sources · 35% ChineseAs of 31 May 2026
This is an evidence-first study that ends in a weighed reading — leans, confidence levels and tripwires for each decisive question — not advocacy for or against Unitree.
Method
Research proceeded by fan-out web search across ~10 question areas in both English and Chinese, followed by direct fetching of primary and reputable secondary sources — Unitree's audited STAR Market prospectus, founder interviews, and the domestic bull/bear debate alongside English-language coverage. Every URL cited was opened and read, not just linked; a mandatory native-language pass meant 35% of sources are Chinese-language, with figures reconciled against the original digits (1亿 = ¥100M; ¥17.08亿 = ¥1.708B). Deliberate disconfirming searches in both languages — criticism, controversy, security, military, bubble — ensured each section carries both sides. Claims were transcribed into a structured manifest tagging each source with a tier, a confidence level and a stance, and the resulting mix (supporting / critical / neutral and language share) is machine-verified for balance with every cited link checked. The load-bearing figures for Unitree are the disclosed prospectus numbers — 2025 revenue, the ~60% gross margin, non-GAAP vs. GAAP profit, ownership, and the ~¥42B (~$6.2B) IPO target — against the contested third-party shipment rankings and the market-size forecasts.
Frameworks used
The study applies the Pyramid Principle for answer-first, neutral synthesis, Porter's Five Forces to map the legged- and humanoid-robotics industry, a peer-comparables benchmark against rivals, a value-chain view to test whether Unitree's in-house components form a durable moat, and a 2×2 price-vs-capability positioning map. Each was applied only where the underlying data supported a real conclusion and was used even-handedly; no framework was forced onto thin data, and forward forecasts are treated as illustrative rather than as a framework output.
Disclosed vs. estimated
Disclosed, high-confidence figures — revenue, profit, gross margin, ownership and the IPO raise — come directly from Unitree's audited STAR Market prospectus and are marked as such. Where figures are stated on a comparable or directional basis, such as non-GAAP profit reconciled to GAAP (~¥600M headline vs. ~¥288M GAAP for 2025), the basis is named in-line. Everything else — private-round valuations, market-share and shipment figures (Omdia/IDC/SemiAnalysis, themselves disputed by Unitree), and all forward forecasts — is a third-party estimate, labeled and attributed rather than presented as fact.
⚠️
Where this case study may be wrong
Shipment rankings are contested. Omdia/IDC rank AgiBot first and Unitree second in 2025 humanoids; Unitree calls those figures "false information" and claims No.1. We show both.
Profit figures differ by basis. The headline ¥600M (2025) is non-GAAP; GAAP was ~¥288M.
Valuations and forecasts are estimates. The ~¥42B IPO figure is a target; market-size forecasts span orders of magnitude.
Critical claims are attributed, not adjudicated. Security and military findings reflect what researchers/lawmakers/outlets reported, with Unitree's responses alongside; we did not independently audit the robots or the supply chain.
It will go stale. The IPO hearing was set for 1 June 2026; outcomes after our as-of date are not reflected.
Neutrality & independence
Each section pairs the case for and the case against, and the Forward View states where the evidence leans, at what confidence, and what would flip each reading — weighed conclusions rather than advocacy. It is an independent research artifact, not affiliated with, authorized by, or endorsed by Unitree Robotics or any company mentioned, and it is point-in-time as of 31 May 2026 — later developments, including the IPO hearing set for 1 June 2026, are not reflected. Trademarks belong to their owners. Nothing here is investment advice — no rating, price target, or recommendation to buy or sell any security.
Bibliography
Sources
Every cited source was fetched during the research run and cross-checked by an automated link checker. Tiers: 1 = primary/official, 2 = reputable press, 3 = forums/sentiment/tertiary.
Unitree (Hangzhou Yushu Technology) was founded Aug 2016 in a 50 m² Binjiang office by Wang Xingxing; ~500 employees by 2025; filed STAR Market IPO March 2026.
Wang Xingxing was the only post-90s entrepreneur to speak at Xi Jinping's 17 Feb 2025 private-enterprise symposium, alongside Ren Zhengfei and Wang Chuanfu.
Wang's design philosophy: relentless focus and that design — not mass production — is the main cost-reduction lever; H1 built by a small team in ~6 months.
Unitree is grouped among Hangzhou's 'Six Little Dragons' (六小龙) with DeepSeek, Game Science, DEEP Robotics, BrainCo and Manycore — a label popularized in early 2025.
Hangzhou's Dec 2024 policy offers up to ¥5M per research project; Unitree disclosed ¥76M of tax incentives in 9M-2025 and ¥32M of grants 2022–Sep 2025.
MERICS: China controls ~63% of the global humanoid-component supply chain and ~90% of permanent-magnet processing; Chinese firms shipped ~90% of 2025 humanoid units.
Chinese investors warn of a bubble: VC Zhu Xiaohu flagged 'valuations that mortgage the future, unclear scenarios, and insufficient technological maturity'.
H1 set a claimed full-size-humanoid speed record of 3.3 m/s (Mar 2024); the record is category-specific and Unitree-claimed, not independently verified.
Chinese reporting and even rival CEO He Xiaopeng note Unitree demos lean on human teleoperation; open-task AI latency exceeds ~5 seconds, and teleoperation is used mainly to collect training data.
Customer concentration is low: top-5 customers were only 10.61% of 9M-2025 revenue, largest (JD) 3.54%; humanoid ASP fell from ¥593K (2023) to ¥168K (9M-2025).
Economic Observer's read of the prospectus: only ~2.6% of humanoid sales (¥15.7M) came from real industrial deployment in 9M-2025; 73.6% was research/education.
The 2025 shipment rankings are genuinely disputed: Omdia and IDC both rank AgiBot first and Unitree second, while Unitree calls the circulating figures 'false information' and reports 5,500+ delivered / 6,500+ produced.
Chinese analysts argue Unitree is 'strong in body, weak in brain' — its moat is gait control, with shallow semantic understanding and weak open-domain generalization versus AI-first rivals.
A key strategic vulnerability: TrendForce notes Unitree's AI models still struggle to generalize across environments — the software, not the hardware, is the gap.
Kharon: ~20% of Unitree's inputs rely on imported materials (Nvidia Jetson, Intel RealSense) via intermediaries — a sanctions-exposed dependency unmentioned in the IPO filing.
The IPO earmarks 85% of the ¥4.2B raise for R&D, including ¥2.02B (48%) for embodied-AI large-model work — an explicit pivot from hardware maker toward a full-stack AI+hardware platform.
Figure AI raised $1B+ at a $39B valuation (Sep 2025) and ran a real BMW deployment — far higher valuation and AI depth than Unitree, at a fraction of the volume.
Xpeng's next-gen Iron humanoid (82 DOF, 3,000 TOPS, solid-state battery) targets mass production by end-2026 at an estimated premium price (~$150K class).
Prospectus revenue: ¥123M (2022), ¥159M (2023), ¥392M (2024), ¥1.167B (9M-2025); net result swung from losses to ¥94.5M profit (2024) and ¥105M (9M-2025).
Full-year 2025 (prospectus estimate): revenue ¥1.708B (+335%); non-GAAP net profit >¥600M (+674%). GAAP net profit was ¥288M; the gap is ~¥349M of stock-based comp.
Unitree was selected for a CSRC on-site inspection in Apr 2026 — routine pre-listing diligence — and, unlike one prior pick, did not withdraw its application.
Wang (68.78% voting control) voluntarily disclosed the weak Q1-2026 profit before the IPO hearing — read as screening for investors who tolerate earnings volatility while he concentrates spend on AI.
Researchers disclosed CVE-2025-2894 — an undocumented CloudSail 'backdoor' in the Go1 enabling remote access to ~1,919 devices, including at MIT and CMU.
Unitree responded that attackers had illegally obtained a third-party tunnel's admin key, called the feature 'common', and shut the service down by 29 Mar 2025.
All 24 members of the US House Select Committee on the CCP urged (May 2025) adding Unitree to military/Entity/FCC lists, citing ties to China's military and restricted entities.
PLA-adjacent uses are documented: a rifle-equipped Go2 in propaganda and B2 robots in the 2024 China–Cambodia 'Golden Dragon' drill (shown on CCTV/Axios).
Unitree's rebuttal: it makes civilian products, any weaponization is by third parties, and it signed a 2022 anti-weaponization pledge; its IPO filing omits military uses.
Kharon documents Unitree sales to Entity-List universities and a partner network tied to a PLA 'Robot Wolf' combat platform; Unitree did not respond to questions.
A Unitree H1 flailed and nearly struck handlers during a factory test (May 2025), raising safety questions; Unitree did not immediately explain the cause.
VC Zhu Xiaohu (GSR) questioned commercialization — 'every humanoid robot can do somersaults, but where is the commercialization?' — advising early investors to exit.
A US 'American Security Robotics Act' (2026) and an FY27 NDAA push would restrict federal use of Chinese ground robots — a direct threat to Unitree's US sales.
Despite the politics, Unitree is already selling into the US (e.g., US Army and police buyers) at a fraction of Boston Dynamics' price, because few American options exist — a fact that cuts both ways.
Per its own prospectus, Unitree had not yet deployed its self-developed general embodied large model at scale in shipping products during the reporting period.
VC Zhu Xiaohu, after exiting early embodied-AI bets, said the sector's heat is 'far too high' and will pass through a bubble and a 'valley of death' before recovering.
Zhu Xiaohu called many embodied-AI customers 'imaginary' and labelled Unitree a 'momentum' investment rather than a fundamentals one — even while expecting its STAR listing to do well.
Experts caution most humanoids today 'dance, flip, and patrol exhibition halls', with new buyers often state firms placing them in lobbies for display.
Regulatory cooling: a Jan 2026 rumor that Unitree's fast-track IPO channel was halted (which Unitree denied applying for) was read as a signal of robotics-sector froth.
Wang predicted humanoids — especially in China — will run a sub-10-second 100m in 2026, framing 2026 as the year robots move from labs to real deployment.
A few primary sources (e.g. paywalled or bot-walled pages) may not resolve for the automated checker and were verified manually. See Methodology & Limits.