DexForce

1.1 Identity, headquarters, founding, and business model

DexForce Technology Co., Ltd. (跨维（深圳）智能数字科技有限公司, also written Kuawei Intelligence or Trans-Dimensional Intelligence in English-language coverage) is a Chinese national high-tech enterprise centered on embodied intelligence and humanoid robots. The company's own English and Chinese about-us pages state it was founded in June 2021, and Baidu Baike's corporate profile gives a more precise incorporation date of June 15, 2021. Headquarters sit at the 14th Floor, Shenzhen Innovation Investment Mansion (also translated Venture Building), No. 1001 Keyuan Road, Nanshan District, Shenzhen, Guangdong Province — an address repeated consistently across the company's English site, Chinese site, documentation center, and the dexforce.top vision-hardware storefront. The company says more than 70% of its staff work in R&D, led by what it calls 'top 2% global scientists.' Its product matrix spans EmbodiChain (an open-sourced, GPU-accelerated embodied-AI development platform published on GitHub under an Apache-2.0 license), the DexForce W1 Pro general-purpose wheeled humanoid, DexSense pure-vision spatial sensors and 3D industrial cameras (marketed in part through the dexforce.top storefront), the DexVerse simulation/synthetic-data engine, and the DexWorldModel world model. Commercially, DexForce blends outright hardware sales with a self-described 'Physical Token' economics model — charging per physical action a robot performs, echoing per-token pricing for large language models — and a franchise-like 'DexBot Inside' operator program in which local partners run robot-staffed retail and cultural-tourism kiosks under revenue-sharing arrangements. The result is a hybrid business model: selling robots and vision hardware outright, licensing the EmbodiChain/DexVerse software stack, and running an operator network that pays for deployed robot capacity rather than a single up-front purchase.[CO001, CO002, CO003, CO004, CO005, CO006]

1.2 Leadership, governance, and key-person dependence

The public leadership narrative centers almost entirely on Professor Jia Kui (贾奎), credited across the company's own news releases and independent coverage as the founder and lead scientist. EmbodiedGlobal's reporting on the Series B describes him as a tenured professor at the Chinese University of Hong Kong, Shenzhen (CUHK-Shenzhen) and a multi-year Stanford 'Top 2%' cited scientist, and DexForce's own Series B announcement quotes him directly on the company's 'highest ROI per physical token' strategy. The company's hiring page adds that the broader core team draws from MIT, Nanyang Technological University, the National University of Singapore, KAIST, Tsinghua University, and Zhejiang University, but names no other individual executives, and no independent board roster, committee structure, or voting-control schedule appears in any source reviewed for this chapter. A notable wrinkle: Baidu Baike's corporate-registry summary lists the company's legal representative as Jia Jun (贾军), a name distinct from founder Jia Kui (贾奎) in the reviewed romanization — a discrepancy that may reflect a separate compliance-facing officer, a transliteration inconsistency across sources, or an unresolved identity question the public record does not settle. Either way, the founder-facing narrative and the registered legal representative are not obviously the same disclosed individual, which compounds the already-high key-person dependence risk typical of a scientist-led deep-tech startup: strategy, fundraising messaging, and technical direction all run through one named professor with no publicly identified deputy or succession plan.[CO011, CO012, CO013, CO014, CO015, CO016]

1.3 Funding history, investor base, and valuation

DexForce's disclosed financing history spans at least three distinct events. In May 2024, EqualOcean reported a strategic financing round led by Lenovo Capital, earmarked for product development, team expansion, and market growth — the earliest third-party-confirmed capital event found. In July 2025, Chinese outlets 21jingji, Investor's China (投资家网), and Chuangtouzhijia reported that DexForce closed combined A1 and A2 rounds worth several hundred million RMB (数亿元), led by Chengdu Sci-Tech Innovation Investment (成都科创投) and Hongtai Fund (洪泰基金), with Tianying Capital, the Sichuan Academician Fund, Nanshan Strategic Emerging Industries Investment, Yicun Songling, and Tanyuan Venture Capital also participating, while existing shareholder Lenovo Capital added further follow-on capital. Then, on June 30, 2026, DexForce announced a Series B of RMB 1 billion (roughly US$137-138 million) at a post-money valuation above RMB 10 billion (roughly US$1.37-1.38 billion), crossing the unicorn threshold. Series B investors named across the company's own release, Gasgoo, Pandaily, EmbodiedGlobal, and Wedoany include returning double-down backers Shenzhen Capital Group (深创投) and the Guiyang Digital Economy Fund; new investors Qianhai Mother Fund, Lens Technology (SZ:300433, which reportedly became a customer before investing), ICBC Capital, Hengjian Capital/Asset, and Zhurui Capital; and follow-on capital from existing shareholders Nanshan Strategic Emerging Industries Investment, Chengdu Sci-Tech Innovation Investment, and the Sichuan Academician Fund. No source reviewed reconciles a full financing ledger — exact dates, tranche sizes, instrument type, or cumulative lifetime proceeds are not public, so total capital raised is best treated as unresolved even though the individual round events are independently corroborated.[CO017, CO018, CO019, CO020, CO021, CO022]

1.4 Scale claims, revenue signals, and commercial milestones

DexForce's scale claims evolve depending on which source and date is checked, and the chapter treats them as company-claimed rather than independently audited. The English homepage (about.html) states deployment across '50+ sub-industries and 1,000+ projects,' while the Chinese about-us page and the company's own June 2026 Series B article state '50+ 细分行业' and '1,500+ 项目' (50+ sub-industries, 1,500+ projects) alongside a separate claim of '1,500+ 具身模型' (1,500+ embodied models) — a similar-looking number that actually describes a different metric (deployed models, not deployed projects), and the English and Chinese project counts themselves do not match, most likely reflecting an update lag between the two language sites rather than two independently verified figures. On revenue, EmbodiedGlobal quotes founder Jia Kui saying DexForce generated more than RMB 100 million in 2025 and booked roughly RMB 100 million in H1 2026 alone, guiding toward 3-4x revenue growth for full-year 2026 (implying a roughly RMB 300-400 million target); separately, an investor quoted in DexForce's own Series B article projects full-year revenue 'in the hundreds of millions of RMB' without committing to a specific figure, and an earlier July 2025 21jingji article already described the company as having reached '亿元级' (RMB 100 million-plus) annualized revenue and being deployed across 'dozens of scenarios' with '100+ enterprise customers' at that time — evidence of a steadily rising, but self-reported, revenue trajectory rather than an audited one. On product milestones, DexForce's DexWorldModel topped the Track 2 (Functionality) leaderboard of the WorldArena global embodied world-model benchmark in April 2026, and the company published its Dexterity-BEV data-alignment method (arXiv 2606.02274) the same period. Commercially, the company moved from trade-show demonstrations toward paid deployment: a May 1, 2026 rollout placed 10+ robot-staffed retail kiosks across five city landmarks in a single day, drawing more than 100,000 visitors, and a same-month internal report claimed per-store holiday revenue above RMB 100,000, a 4.6x revenue lift versus traditional staffed kiosks, 99% store-visit-to-purchase conversion, and payback periods as fast as six months. A May 21-25, 2026 appearance at the China (Shenzhen) International Cultural Industries Fair produced national broadcaster (CCTV) coverage and new operator-partner signings under the 'DexBot Inside' program.[CO029, CO030, CO031, CO032, CO033, CO034]

1.5 Adverse context and underwriting limits

The strongest counterweight to DexForce's narrative comes from EmbodiedGlobal's own reporting on the Series B, which explicitly cautions that 'like many Chinese unicorns, KuaWei's numbers (1,500 models, 99.9% success, sub-mm precision) are self-reported and should be treated as claims until third-party benchmarks are available,' and that the jump to a RMB 10 billion-plus valuation 'will need commensurate revenue growth through the IPO window.' That caution sits inside a broader sector-level warning: Morgan Stanley's November 2025 research note (covered by ChinaBizInsider) forecasts a far more conservative 2026 China humanoid-robot shipment volume (about 12,000 units) than industry bulls project (some over 100,000), citing current robot task efficiency at only 20-30% of a human worker and describing near-term commercial deployment as gated by cost and reliability rather than hype. While that note is not DexForce-specific, it applies directly to any diligence built on DexForce's self-reported 99.9%+ task-success and rapid-payback retail claims. Beyond scale-figure skepticism, the chapter surfaces several disclosure gaps an investor would need closed before underwriting the post-Series-B valuation: no audited financial statements, no disclosed headcount, no public cap table or voting-rights schedule, and no customer-concentration breakdown despite named enterprise customers (Midea, CRRC, Wuling, Lens Technology) suggesting meaningful commercial-customer overlap risk. The unresolved legal-representative-versus-founder naming discrepancy adds a further governance-transparency question that later chapters (leadership, financials, risks) should track rather than resolve here.[CO043, CO044, CO045, CO046, CO053, CO054]

1.6 Exhibits

2.1 Market Boundary and Status-Quo Substitutes

DexForce is best understood not as a single-product humanoid maker but as a generalized embodied-AI 'brain' company that licenses a common perception, simulation, and control stack (DexVerse, DexWorldModel, DexAgent, DexSense, DexBrain) across two distinct commercial motions: industrial embodied-AI deployment for flexible sorting, assembly, and material handling across auto parts, new-energy vehicles, 3C electronics, aerospace, logistics, and chemicals, and a commercial-service kiosk business that puts W1 Pro humanoids into convenience stores, malls, theme parks, and tourism landmarks to prepare food, sell retail goods, and interact with visitors. A third, smaller adjacency is DexForce's legacy machine-vision/industrial-3D-camera line, which serves a related but distinct quality-control buyer. The market boundary therefore includes embodied-AI hardware, foundation-model/software licensing, and robot-as-a-service kiosk operations, and excludes traditional fixed-purpose industrial robot arms and cobots that lack embodied generalization, generic factory-automation capex, and consumer/education robot toys with no embodied-AI engine. The status-quo substitute for most of DexForce's target tasks is simply human labor: line workers and process engineers in manufacturing, and staffed retail/food-service employees in the commercial-kiosk segment. DexForce also competes for the same budgets as more than 140 domestic humanoid manufacturers that released over 330 models in 2025, including Unitree and AgiBot, whose 2025 disclosures show humanoid revenue only recently overtook quadruped-robot revenue for the segment leader -- a reminder that even the most commercially advanced peer's humanoid-specific revenue base is very young.[CM001, CM002, CM003, CM004, CM005, CM006]

2.2 Sizing Lenses and the TAM/SAM/SOM Contradiction

Public sizing for China's humanoid/embodied-AI market is unusually contested, and the contradictions are informative rather than noise. Morgan Stanley's latest (June 2026) note pegs China's external-sales humanoid market at $2 billion in 2026, growing to $15 billion by 2030 on 446,000 annual unit shipments -- but that $2 billion lens excludes prototypes, pre-orders, and internal-use units, and it has already been revised sharply: the same bank projected only 12,000 2026 units in a November 2025 'reality check' note, then 14,000 in January 2026, then 28,000, before settling on 50,000 in June 2026 -- a more than four-fold swing in shipment volume from one publisher within seven months. A separate lens from industry tracker GGII, reported via China Daily, counts production output rather than external sales and puts 2025 shipments at 14,400 units with 2026 output potentially reaching 100,000-200,000 units -- two-to-four times Morgan Stanley's shipment number for the same year, because 'output' and 'externally sold shipments' are not the same measurement. An April 2025 conference forecast from consultancy Leaderobot went further still, projecting 82 billion yuan (about $11.6 billion) of 2025 China humanoid revenue alone -- far above what either the GGII shipment data or Morgan Stanley's market-size figure would imply. Regional industrial-cluster figures add a fourth kind of number entirely: Hangzhou's embodied-AI cluster already reported 106.8 billion yuan (about $15.7 billion) of 2025 output value for one city, and Shanghai targets over 50 billion yuan (about $6.96 billion) of core output by 2027 -- output-value figures for regional industrial clusters, not demand-side humanoid-unit market sizes, so they cannot be summed with the national shipment-revenue lens. No source reviewed isolates a bottom-up figure for DexForce's own obtainable share of any of these lenses; its roughly $1.4 billion post-money Series B and 50+ live scenario footprint are the only public anchors for a directional, low-confidence SOM.[CM011, CM012, CM013, CM014, CM015, CM016]

2.3 Buyer, User, and Payer Segmentation and the Adoption Path

DexForce's adoption path runs through at least four distinct buyer motions. In industrial manufacturing, the buyer/payer is the plant-operations or manufacturing-engineering budget owner, often organized into an 'innovation application consortium' alongside integrators, model providers, and research institutes under MIIT and SASAC's 2026 joint real-scenario training action, which directs each province to select at least 20 scenario units and each central state-owned enterprise at least 10, spanning industrial, service, and special domains, with a stated goal of ten-thousand-unit-scale deployment capability by the end of 2026. In the commercial-service kiosk segment, the buyer/payer is a local operator or franchisee that DexForce recruits nationally, while the end user is the walk-in consumer; DexForce's own May Day rollout placed 10+ kiosks across five cities and reported payback periods as fast as 6 months, single-store holiday revenue above 100,000 yuan, a 4.6x revenue uplift versus traditional staffed stores, and footfall increases of 30%-40% (up to 200% at flagship sites) -- figures that remain company-disclosed and are not yet independently audited. A third path runs through government- and SOE-anchored demonstration sites, where the trigger is meeting a policy quota rather than a conventional purchase decision, and Shanghai's implementation plan shows one concrete example of a co-payer structure, funding qualifying projects up to 50% of cost (capped at 20 million yuan) and pilot deployments up to 20% of investment (capped at 10 million yuan); the national MIIT/SASAC notice does not disclose an equivalent funding percentage, leaving that program's payer economics only partially observable. Across every segment, adoption converts only after a documented application-verification step -- MIIT's own language is to 'validate one scenario, deploy a batch, and drive a cluster' -- rather than a single enterprise sales cycle.[CM029, CM030, CM031, CM032, CM033, CM034]

2.4 Growth Drivers and Adoption Constraints

Growth drivers are converging from policy, cost, and demand simultaneously. Beijing has named embodied AI a five-year strategic priority, directing subsidized land/office space and favorable bank lending; MIIT and SASAC's 2026 real-scenario training action gives that priority a concrete, quota-driven deployment path; China's first national standard system for humanoid robots and embodied AI, spanning safety, application, and lifecycle-management pillars, was released in March 2026 after collaboration among more than 120 institutions, reducing integration and compliance uncertainty; and Hangzhou's binding local regulation, effective May 2026, is the first legally defined framework for the sector and could become a template other provinces copy. On the cost side, falling average selling prices (Unitree's own ASP fell roughly 72% from 2023 to the first three quarters of 2025) and greater in-house component production, leveraging shared supply chains with China's new-energy-vehicle industry, are improving buyer ROI economics across both industrial and commercial deployments. TrendForce reports growing 2026 orders from automotive manufacturing, consumer electronics, and logistics as trial adoption shifts to real demand. The constraints are just as material. Morgan Stanley's own November 2025 note flagged 'limited working capability' -- brain/VLA model and training-data gaps, body-component quality-consistency issues, and price/cost pressure -- as the binding constraint on near-term adoption, and MIIT's own 2026 notice explicitly calls for building out collision-detection, force-control, emergency-braking, and safety-recording capability that is not yet mature. A Beijing NDRC official and Goldman Sachs have both separately warned of bubble and overcapacity risk given more than 150 Chinese humanoid companies (over half startups or cross-industry entrants) and production scaling ahead of confirmed order volume. Regulatory maturity is also uneven: only Hangzhou has binding local legislation so far, while most other regions rely on non-binding strategic plans or subsidized pilot programs, and intense global competition from Tesla Optimus, Boston Dynamics Atlas, Figure AI, and 1X, alongside domestic rivals Unitree, AgiBot, and UBTECH, adds pricing and execution pressure even for well-funded Chinese players such as DexForce.[CM039, CM040, CM041, CM042, CM043, CM044]

2.5 Exhibits

3.1 Framing the Landscape: Direct Peers, Incumbents, Adjacents, Substitutes, and Internal Build

DexForce sits inside a crowded, regulator-flagged Chinese humanoid and embodied-intelligence sector. A Beijing NDRC official said in November 2025 that more than 150 Chinese companies were making humanoid robots, with over half being startups or cross-industry entrants, and warned of bubble and overcapacity risk from repetitive, undifferentiated products. Within that population, this chapter treats five companies as direct peers on capability and go-to-market grounds -- Unitree Robotics, AgiBot, UBTECH Robotics, Fourier Intelligence, and Robotera -- because each ships or is developing a general-purpose humanoid or embodied-intelligence platform for industrial, commercial, or consumer deployment. Boston Dynamics' Atlas (already in commercial deployment) and Tesla's planned Optimus Gen 3 mass production sit one layer out as adjacent global incumbents that TrendForce explicitly tracks against the China cohort. Two further layers matter for DexForce specifically: status-quo substitutes -- traditional fixed-position industrial automation and human staffing, both of which DexForce's own retail kiosks and industrial deployments are trying to displace -- and internal build, since large industrial customers such as Midea, CRRC, and Wuling could in principle develop in-house robotics/automation capability rather than licensing DexForce's models. Likely entrants include the wave of sub-RMB30,000 consumer humanoid makers (Noetix, Booster, Astribot) that China Daily reports are pushing prices down industry-wide, a commoditization dynamic that could eventually reach DexForce's higher-priced wheeled-humanoid category.[CP047, CP037, CP038, CP052, CP053, CP039]

3.2 Direct Peer Profiles: Scale, Funding, Target Customer, and Strategic Direction

Unitree Robotics (Hangzhou) is the scale leader: its STAR Market IPO filing discloses RMB1.7 billion 2025 revenue, RMB600 million adjusted net profit, 60.27% gross margin, and 5,500+ humanoid units shipped in 2025 (#1 globally), against a RMB12.7 billion pre-IPO valuation and a targeted RMB42 billion minimum IPO market cap; it self-manufactures over 90% of core components and plans to expand to 75,000 humanoid and 115,000 quadruped annual capacity. AgiBot (Shanghai) reached its 15,000th mass-produced general-purpose robot in June 2026, having scaled from 1,000 to 10,000 units in three months, spans the A2, G1, X1, and X2 product lines, and set a Guinness World Record for humanoid walking distance; TrendForce credits its standardized, order-driven supply chain for the ramp. UBTECH Robotics is the only Hong Kong-listed (9880) pure-play humanoid company in this set, targeting factory assembly lines with Walker S/S2 (the latter the first humanoid with autonomous battery swapping) alongside a legacy AI-education business. Fourier Intelligence (Shanghai) positions its GRx series and RehabHub product toward healthcare/rehabilitation and consumer-companion use cases rather than industrial deployment, a softer target market than DexForce's. Robotera (Beijing) markets itself as the industry's only fully self-developed software-and-hardware embodied-intelligence stack, including the XHAND1 dexterous hand, though its official product pages disclose little specification or pricing detail. DexForce itself reports more than RMB100 million 2025 revenue, guidance for 3-4x 2026 growth, a RMB10 billion+ Series B valuation, and named industrial customers (Midea, CRRC, Wuling) -- a materially smaller and less transparent scale footprint than Unitree's or AgiBot's disclosed production and financial figures.[CP015, CP016, CP017, CP018, CP021, CP023]

3.3 Capability, Pricing, Go-to-Market, and Trust/Regulatory Comparison

On disclosed capability, DexForce's W1 Pro claims sub-millimeter arm precision (±0.5mm) and up to 40 degrees of freedom with a 10kg single-arm payload -- figures none of the profiled peers publish at comparable granularity -- but it is a wheeled, not bipedal, platform, while Unitree, AgiBot, UBTECH, Fourier, and Robotera all field bipedal humanoids, and independent coverage notes that Fourier, UBTECH, and AgiBot are all investing heavily in VLA (vision-language-action) model capability alongside Unitree. On pricing, Unitree is the only competitor with fully public list pricing (G1 from US$13,500; R1 dual-arm from RMB26,900/~$3,968), while AgiBot, UBTECH, Fourier, and Robotera all direct buyers to sales contact, matching DexForce's own lack of published pricing (third parties estimate the W1 Pro at $50,000-$150,000, or roughly $78,000). On go-to-market, DexForce's franchise-style "robot commercial service kiosk" operator-recruitment model for retail/tourism venues is structurally distinct from every peer's direct industrial/enterprise or consumer-retail sales motion. On trust and regulatory posture, China released its first national humanoid-robot standard system (HEIS, "2026 Edition") in February 2026, and the drafting committee was reported to include the founders of Unitree and AgiBot, giving those two companies a direct voice in standard-setting that DexForce is not reported to hold; separately, MIIT and China's State Council SASAC jointly launched a 2026 real-scene training program to validate humanoid deployment across industrial, service, and specialized scenarios nationally.[CP007, CP008, CP009, CP054, CP019, CP020]

3.4 Switching Cost, Lock-In, Distribution Power, and Supply/Partner Access

DexForce's retail-kiosk channel is architected for low switching cost and fast operator payback -- the company reports a single-store revenue above RMB100,000 during the 2026 May Day holiday and a payback period as fast as six months -- which lowers the barrier for operator-partners to adopt but by the same logic lowers the barrier for a rival hardware maker to displace DexForce inside that same operator relationship once a cheaper or more capable robot is available. On the industrial side, DexForce's disclosed customers (Midea, CRRC, Wuling) reflect AI-model licensing and deployment rather than confirmed exclusive long-term contracts in the sources reviewed, which leaves switching cost and lock-in largely undocumented. Distribution power diverges sharply across the peer set: UBTECH's Hong Kong listing gives it direct capital-markets access and disclosure obligations DexForce does not yet have as a private, IPO-preparing company; Unitree's pending STAR Market listing would extend the same advantage; and AgiBot's order-driven, standardized manufacturing supply chain is credited by TrendForce with enabling its rapid 2026 production scale-up in a way DexForce has not disclosed matching. On supply/partner access, Unitree self-manufactures over 90% of core hardware components (motors, reducers, lidar), giving it direct control over cost and supply during an industry-wide period of falling average selling prices; DexForce has not disclosed a comparable vertical-integration position, leaving its hardware cost structure and component supply chain as an open diligence question relative to peers that either manufacture in-house (Unitree) or have engineered a standardized external supply chain (AgiBot).[CP012, CP005, CP051, CP021, CP025, CP055]

3.5 Moat Durability, Commoditization Risk, and Adverse Evidence

DexForce's central moat claim is generative-simulation AI: the DexVerse engine and DexWorldModel (ranked #1 on the WorldArena embodied world-model benchmark) let it generate synthetic training data rather than rely on the large-scale teleoperation data collection many Chinese peers use. Independent trade coverage of Unitree's own IPO filing directly challenges the durability of that moat, arguing that Unitree's "data flywheel" -- 30,000+ quadrupeds and 5,500+ humanoids already generating real-world interaction data -- is a more durable advantage than "pure-model companies" relying on simulation alone, and that Fourier, UBTECH, and AgiBot are all separately racing toward comparable AI-model capability. Commoditization risk is already visible at the low end: China Daily reports Unitree's own average humanoid selling price fell from RMB593,000 (2023) to RMB167,000 (9M 2025), Leju's Kuavo ASP fell 26% year-over-year to RMB308,100, and sub-RMB30,000 consumer entrants (Noetix's Bumi at RMB9,998, Booster's K1 at RMB29,900) and a sub-RMB100,000 industrial model (Astribot's T1 at RMB89,900) show how fast price points are compressing across the category DexForce's higher-estimated-price W1 Pro also competes in. Adverse, sector-wide evidence reinforces the risk: Morgan Stanley's November 2025 note forecasts only about 12,000 China humanoid units in 2026 (versus some players' 100,000-unit projections), citing current robot efficiency at just 20-30% of human levels, integration complexity, and price/cost pressure as binding constraints; separately, a Beijing NDRC official and Goldman Sachs have both warned of bubble and overcapacity risk across the sector's 150+ companies. For a private, pre-IPO company like DexForce with no disclosed profitability or manufacturing scale, that combination of commoditization pressure and financing-market skepticism is a materially higher risk than for already-profitable (Unitree) or already-listed (UBTECH) peers.[CP013, CP014, CP056, CP022, CP040, CP039]

3.6 Exhibits

4.1 Revenue Streams, Pricing Model, and Recognition

DexForce's own materials describe a multi-stream commercial model rather than a single product line: industrial humanoid licensing/deployment to manufacturing customers (Midea, CRRC, Wuling cited as deployments), a "Physical Token" consumption-based pricing construct for embodied-AI compute/data services, direct hardware sales (e.g., the W1 Pro humanoid platform), and a franchised retail-robotics concept ("DexBot Inside") that shares kiosk revenue with local operators recruited nationwide. Company press releases quote per-store economics (payback period, revenue uplift, peak daily orders) for the DexBot Inside format and describe multi-city rollout activity around May 1, 2026, but none of these figures are independently audited. Aggregate revenue reporting is similarly fragmented and self-reported: Gasgoo cites RMB100 million in H1 2026 revenue against an FY2026 target above RMB300 million; Embodied Global cites 2025 revenue above RMB100 million with 3-4x growth guidance for 2026; Pandaily cites an FY2026 target of RMB250-300 million. No source discloses revenue recognition policy, the split between licensing/subscription vs. hardware-sale revenue, or whether "revenue" figures are booked, contracted, or shipment-implied. The three FY2026 targets do not reconcile to a single figure, and no source clarifies whether the RMB100 million H1 2026 figure is cash-collected, invoiced, or a run-rate estimate. This is treated as a conflicting-data claim rather than resolved.[CI001, CI002, CI003, CI004, CI005, CI006]

4.2 Go-to-Market Motion and Sales-Efficiency Proxies

DexForce's disclosed go-to-market motion blends industrial B2B deployment (named customers Midea, CRRC, Wuling per Embodied Global) with a consumer-facing franchise channel (DexBot Inside kiosks) and public-facing brand exposure (state-media coverage at a cultural-industries expo). The kiosk format is the only channel with quantified unit economics in company materials: single-store monthly revenue above RMB100,000, payback inside a stated window, foot-traffic conversion uplift, and peak daily order counts, all self-reported and none independently verified by a third party. No source discloses customer acquisition cost, sales headcount, sales-cycle length, net revenue retention, or contract duration for the industrial channel, making it impossible to compute standard SaaS/hardware sales-efficiency proxies (e.g., magic number, CAC payback) with any confidence. The shipment cadence implied by "several hundred" H1 2026 humanoid units against a stated near-1,000-unit full-year 2026 target implies material second-half loading, which is consistent with either seasonal industrial procurement cycles or optimistic full-year guidance; no source disaggregates the channel mix (industrial vs. kiosk vs. export) of either figure. Comparable-company sales-efficiency proxies from UBTECH (audited, filing-sourced) and Unitree (IPO-linked prospectus reporting) are presented for contrast rather than direct substitution.[CI011, CI012, CI013, CI014, CI015, CI016]

4.3 Cost Structure, Gross-Margin Drivers, and Capital Expenditure

DexForce discloses no gross margin, cost of goods sold, or capital expenditure figures in any located source. Its own hardware specifications (dual 7-DOF arms rated 10kg payload each at ±0.5mm precision, dual onboard computers rated at or above 300 TOPS, 8+ hours of wired continuous operation) imply a cost structure dominated by precision actuation, compute silicon, and sensing components consistent with the broader humanoid-robotics category, but no bill-of-materials or supplier cost data is available to quantify this for DexForce specifically. In the absence of DexForce-specific disclosure, this chapter triangulates plausible margin and capex ranges from two public comparables: UBTECH's audited FY2025 gross margin of 37.7% (up from 28.7% in FY2024) on consolidated revenue of RMB2,001.0 million, with its newly scaled full-size humanoid segment growing revenue 2,203.7% year-over-year to RMB820.6 million on 1,079 units; and Unitree's reported 60.27% gross margin (up from 44.22% in 2023) on 2025 revenue of approximately RMB1.7 billion and 5,500+ humanoid shipments, alongside a third-party actuation-cost estimate (McKinsey, cited by The Robot Report) that bill-of-materials costs represent 40-60% of humanoid hardware cost. These figures should not be read as DexForce's own margins; they establish a plausible range (roughly 30-60% gross margin) for a scaling Chinese humanoid producer, bounded by hardware cost intensity on the low end and software/licensing mix on the high end, that DexForce has not confirmed applies to its own business.[CI022, CI023, CI024, CI025, CI026, CI027]

4.4 Public Traction vs. Private Financial Gaps

DexForce is a privately held company with no located audited financial statements, annual report, or regulatory filing. Corporate-registry lookups attempted directly (qcc.com, cnverify.com, fundup.ai) each returned access-blocked responses (404 or Cloudflare bot-check pages), preventing independent verification of registered capital, shareholder structure, or filed financial summaries beyond what Baidu Baike reports (registered capital RMB1,557,874 as of November 2024). That registered-capital figure is more than two orders of magnitude smaller than the cumulative funding reported across the A1/A2 rounds and RMB1 billion Series B, which is unremarkable for a Chinese company structure (registered capital is a nominal balance-sheet figure, not paid-in cash) but underscores that no source ties disclosed fundraising amounts to an audited balance sheet. Public-market comparables illustrate what disclosure could look like at scale: UBTECH's HKEX-filed 2025 annual report gives line-item revenue, gross profit, and net loss (RMB789.8 million) figures, and its market capitalization (HK$54.73 billion as of July 2-3, 2026, down from a HK$63.48 billion year-end-2025 peak) is a continuously observable proxy for investor sentiment; Unitree's IPO-linked reporting discloses revenue, margin, and channel-mix detail unavailable for DexForce. The gap between DexForce's press-release-level disclosure and these public comparables is the central financial-diligence constraint of this chapter.[CI029, CI030, CI031, CI032, CI033, CI034]

4.5 Capital Adequacy and Financing Dependency

DexForce's disclosed financing history includes an undisclosed-amount 2024 round involving Lenovo Capital, A1/A2 rounds in 2025 reported at "hundreds of millions of yuan" (Baidu Baike, Chuangtouzhijia) with a named investor syndicate (Chengdu Sci-Tech Innovation, Hongtai Fund, Tianying Capital, Sichuan Academician Fund, Nanshan Strategic Emerging, Yicun Songling, Tanyuan Venture Capital, and a Lenovo Capital follow-on), and a RMB1 billion Series B in 2026 described by DexForce's own site as reaching a valuation in the tens of billions of RMB. Historical (2025-vintage, now unreachable) cached coverage from 21jingji and InvestorsCN quotes an unnamed Lenovo Capital investor stating DexForce had "100+ cumulative customers, dozens of scenarios, and annual revenue at the hundred-million-RMB tier" as of mid-2025 — an early independent-investor data point that predates and is broadly consistent with later 2025/2026 revenue reporting, though both source URLs now return 404 and could not be re-verified live. No source discloses cash on hand, monthly cash burn, runway, total debt, or whether DexForce is presently profitable or cash-flow positive; given a hardware- and services-led business model, reliance on successive equity rounds to fund capital expenditure and working capital appears likely but is not confirmed. This is a material and, on cash-runway specifically, a blocking evidence gap for underwriting purposes.[CI036, CI037, CI038, CI039, CI040, CI041]

4.6 Commercialization Skepticism: An Adverse Read

The Robot Report's analysis of Unitree's IPO disclosure provides an adverse-stance check on humanoid-robotics commercialization narratives generally, which is directly relevant to reading DexForce's own claims with appropriate skepticism. Despite Unitree's headline revenue and margin growth, the outlet finds that only 9.01% of Unitree's humanoid-segment revenue in the nine months to September 2025 came from industrial applications, versus 73.6% from research/education customers and 17.4% from commercial demonstrations — implying that even the sector's most commercially advanced, IPO-bound player derives the large majority of humanoid revenue from non-industrial use cases. The same analysis notes a declining R&D-to-revenue ratio (31.39% to 7.73%) and cites a McKinsey estimate that actuation and sensing bill-of-materials can represent 40-60% of humanoid hardware cost, both consistent with a business still working out durable unit economics at scale. Because DexForce discloses no customer-segment revenue split, it is impossible to determine whether its own Midea/CRRC/Wuling industrial references represent revenue-generating deployments, pilots, or demonstration engagements; the Unitree finding is a reasonable basis to treat DexForce's "industrial customer" framing as unverified until a comparable segment disclosure is available. This does not invalidate DexForce's reported revenue growth, but it materially discounts confidence that the growth reflects sustainable industrial-automation demand rather than pilot, research, or demonstration activity typical of the sector at this stage.[CI029, CI043, CI044]

4.7 Financial Verdict

Revenue quality: DexForce's reported revenue growth (RMB100 million-plus on a 2025 or H1 2026 basis, with FY2026 targets of RMB250-400 million depending on source) is directionally credible given multiple independent outlets and an early independent-investor quote corroborating a hundred-million-RMB revenue tier by mid-2025, but the absence of a single reconciled figure, a disclosed customer/segment mix, or any audited statement means revenue quality cannot be assessed with confidence — the sector-wide adverse finding on Unitree's low industrial-revenue share is a specific reason for caution. Margin path: no DexForce-specific gross margin is disclosed; comparable Chinese humanoid producers show margins improving from the high-20s/low-30s percent toward the 40-60% range as volume scales, suggesting a plausible but unconfirmed path for DexForce. Capital intensity: the business combines hardware capex (arms, compute, sensors) with a franchise/services layer (DexBot Inside) that may be less capital-intensive per unit of revenue than pure hardware sale, but no capex or unit-cost data is available to confirm this thesis. Blockers to underwriting: (1) no audited financial statements or filings; (2) no disclosed cash balance, burn rate, or runway; (3) three non-reconciled FY2026 revenue targets; (4) no customer-concentration or contract-term disclosure for the DexBot Inside channel or industrial customers; (5) blocked access to Chinese corporate-registry sources that could otherwise independently verify capital structure. Overall, DexForce presents a plausible but financially opaque growth story that would require management-provided audited or reviewed financials, a single reconciled revenue and margin figure, and customer-concentration disclosure before underwriting could proceed with confidence.[CI043, CI044, CI045, CI046, CI047]

4.8 Exhibits

5.1 Product definition in customer workflow terms

DexForce (跨维智能) describes itself as building an integrated hardware-software product matrix — the EmbodiChain embodied-intelligence development platform, the DexForce W1 Pro universal humanoid robot, and DexSense pure-vision spatial-intelligence sensors — deployed, in the company's own words, across 50+ sub-industries and 1,000+ projects. The clearest publicly documented customer workflow is not generic humanoid automation but a scripted commercial-service loop: an operator selects a robot configuration (full-body mobile or stationary half-body), integrates it either through the ROS2-based secondary-development API or a turnkey application such as the "barista" coffee-service package, runs a signed pre-opening equipment/material checklist, and then lets end customers order via voice or a tablet interface on the same local network as the robot. DexForce's own May 2026 release frames this as a generalizable "brain" that can switch between espresso machines, draft-beer taps, and ice-cream/popcorn dispensers across more than ten simultaneously opened storefronts, explicitly contrasting this with single-purpose vending automation. The company also positions each live deployment as a data flywheel: real operating data from commercial-service stations is intended to feed back into DexWorldModel and EmbodiChain training, closing what an internal release calls the "data-model-embodiment-scenario" loop. This is a coherent, workflow-specific narrative, but it rests almost entirely on DexForce's own releases and manuals rather than named third-party operator case studies. [CE001, CE002, CE003, CE022, CE029, CE030]

5.2 Module, SKU, and documentation asset map

DexForce's public surface enumerates at least six distinct product/asset lines. The W1 and W1 Pro humanoid ships in full-body (34 DOF, mobile, laser-SLAM navigation) and stationary half-body (22 DOF) configurations, both documented in parallel English and Chinese spec pages. DexVerse™ and DexWorldModel are the simulation/world-model engines behind policy training; EmbodiChain is the one component DexForce has open-sourced on GitHub under an Apache-2.0 license. RobotToday's independent supplier profile adds two lines not prominent on DexForce's primary marketing pages — a "Kingfisher" 3D smart camera and "PickWiz" vision software — consistent with the separate en.dexforce.top vision-solutions site, suggesting DexSense/vision hardware is sold as a distinct industrial-vision product rather than only as a humanoid accessory. Finally, DexForce's documentation center (docs.dexforce.com) packages a Humanoid User Manual, a Secondary Development API reference, and a Barista Operation Manual, plus separate documentation trees for its 3D Vision Guidance Kit and camera hardware. Notably, third-party listings (RobotToday, RBTX) describe configurations with "up to 40 DOF," while DexForce's own English and Chinese W1 Pro pages list 34 DOF for the full-body unit — a discrepancy no reviewed source reconciles, and one worth flagging before treating either figure as canonical. [CE001, CE010, CE022, CE023, CE034, CE040]

5.3 Architecture: Sim2Real, world model, BEV alignment, and the data engine

DexForce's technical architecture runs from perception and actuation hardware up through a simulation-and-data-engine layer into a world-model/policy layer. On hardware, the W1 Pro pairs a 6cm-baseline binocular pure-vision system with a large-model binocular algorithm, dual 7-DOF arms (10kg payload, ±0.5mm repeatability), and a dual-PC compute architecture rated at ≥300 TOPS running whole-body control at 1kHz. Above that sits DexVerse™, which DexForce says generates physics-accurate 3D scenes and synthetic training data from multimodal inputs using domain randomization and differentiable simulation to reduce the sim-to-real gap; the company also claims resulting models reach "up to 99.9% accuracy," a figure with no independent audit trail. DexWorldModel is positioned as a causal latent world model and, together with EmbodiChain and the Dexterity-BEV research line, forms what DexForce's Core Technology page calls its three flagship technical outputs. DexForce announced that its DSCFuncWorld world model ranked first globally on the WorldArena Track 2 (Functionality/Data-Engine) leaderboard, which it describes as testing whether a world model can generate machine-trainable, transferable, and executable robot data rather than just realistic video. Separately, DexForce published "Dexterity-BEV," a paper proposing a bird's-eye-view alignment layer that normalizes heterogeneous multi-camera, multi-embodiment, multi-operator robot data into one physical coordinate frame — explicitly modeled on the way BEV representations unified autonomous-driving perception. EmbodiChain, DexForce's only openly published component, packages asset generation, scene layout, trajectory sampling, failure recovery, and online data streaming as the data-engine layer feeding this stack, and its GitHub organization shows a commit as recent as July 5, 2026, one day before this report's run date. [CE004, CE005, CE008, CE009, CE011, CE012]

5.4 Deployment, integration, reliability, support, and roadmap

DexForce's clearest deployment evidence is its commercial-service-station rollout: a May 2026 release describes 10+ storefronts opened simultaneously across Xuzhou, Shenzhen, Guilin, and Yangshuo tourism and retail sites, drawing more than 100,000 visitors on launch day, followed by a second release claiming single-store daily revenue above RMB 100,000, a 4.6x revenue lift versus traditionally staffed stores, a 99% visit-to-purchase conversion rate, and a payback period as fast as six months during the Labor Day holiday window. These are entirely self-reported, holiday-peak figures with no named third-party operator audit. Integration support is documented in more procedural detail than most early-stage humanoid vendors publish: the Barista Operation Manual specifies a signed pre-opening checklist, capsule/cup restocking thresholds, and a fault-escalation phone line (400-8086-299) for unresolved errors, while the Humanoid API (v0.4.0) exposes ROS2 topics/services/actions for joint-level state and position control — but only at the low-level joint-control layer in the pages reviewed, with no higher-level task or skill API surfaced. DexForce's own documentation and code are versioned but early: API v0.4.0, W1 user manual v0.4.0, and Barista manual v0.1, with EmbodiChain's GitHub roadmap note stating more features will be added "in the coming months" while the repository remains labeled Alpha. RBTX, a robotics-component marketplace, separately lists a 3-week delivery lead time for W1 Pro units sold through its channel. No source reviewed discloses fleet-level uptime, mean-time-between-failure, or warranty/RMA statistics for any deployed unit. [CE028, CE031, CE032, CE037, CE040, CE041]

5.5 Differentiation, IP/open-source posture, and trust/safety/compliance controls

DexForce's stated differentiation rests on team pedigree (over 70% of headcount in R&D, led by scientists it describes as top 2% globally) and a mixed open/closed IP posture: EmbodiChain is openly published on GitHub under Apache-2.0, while the W1 Pro's control software is characterized as closed-source by an independent marketplace listing, and DexWorldModel weights are not published anywhere reviewed. That partial openness is itself a differentiator versus the broader China humanoid market, where TrendForce forecasts 2026 unit output surging 94% with Unitree and AgiBot alone capturing nearly 80% of shipments — implying DexForce competes on world-model/data-engine specialization and commercial-service niches rather than on hardware unit volume. On trust and safety, the W1 user manual is unusually explicit: it lists prohibited misuse scenarios (passenger transport, medical/life-support use, outdoor operation, steep slopes, overloading, towing, liquid/chemical transport), documents a back-mounted emergency-stop button, an optional pull-cord e-stop, a chassis anti-collision strip, and a battery-management system for the 48V/40Ah pack, and states plainly that the robot has not passed medical-safety certification. No reviewed source — official page, manual, or third-party listing — discloses any independent industrial safety certification (CE, UL, ISO, or equivalent) for the platform, and a separate market commentary reports Morgan Stanley cautioning that China's humanoid-robot commercialization narrative broadly carries "bubble" risk, running ahead of validated field reliability. Taken together, DexForce shows genuine documentation discipline on intended-use and misuse boundaries, but the absence of third-party certification, fleet reliability data, and independently verified benchmark methodology leaves the platform's safety and performance claims resting on the company's own word. [CE025, CE026, CE027, CE034, CE035, CE036]

5.6 Exhibits

6.1 Customer Segmentation

DexForce's disclosed customer base spans at least seven distinct buyer/user/payer segments rather than a single go-to-market motion. On the industrial side, procurement teams at large manufacturers (Midea in home appliances, CRRC in rail equipment, Wuling in automotive assembly, and Lens Technology in precision glass/cover components) buy DexForce's embodied-AI models for bin-picking, assembly, and insertion tasks running on third-party robotic arms; line workers are the day-to-day users while the OEM is the payer. A second, structurally different segment is commercial retail and food-and-beverage service, where DexForce does not sell directly to consumers but instead supplies its "DexBot Inside" embodied-AI brain to independent operators who franchise the hardware into high-traffic venues (theme parks, malls, tourist streets); the operator is both buyer and payer, while consumers are the end users. A third, newly surfaced segment is research and education, evidenced by a single named institutional partner, Guilin Tourism University, alongside three other 2026 partners in Guangzhou and Meizhou whose business models are not detailed in any source fetched this run. Geography is concentrated in mainland China (Shenzhen, Xuzhou, Guilin, Yangshuo, Shanwei, Guangzhou, Meizhou); no source fetched this run documents an overseas commercial deployment. The customer journey map (FU001) traces how a prospect typically moves from trade-show visibility through a pilot, into a named disclosure, and finally into either an industrial production line or a retail franchise signing, with post-sale support and, in Lens Technology's case, an eventual investor relationship.[CU001, CU004, CU005, CU006, CU010, CU011]

6.2 Adoption Trajectory & Deployment Proxies

The clearest independent adoption proxy comes from financial-media coverage (21jingji and InvestorsCN, both dated around July 10, 2025) reporting that DexForce had surpassed 100 paying enterprise customers and RMB100 million-plus in annualized revenue across dozens of deployment scenarios. DexForce's own official pages complicate this picture: the English-language about page claims 50-plus sub-industries and 1,000-plus delivered projects, the Chinese-language variant of the same page claims 1,500-plus, and the careers page separately claims 1,000-plus enterprises served — three different denominators (customers, projects, enterprises) that are never reconciled against one another or against the independently reported 100-customer figure. The most recent deployment proxy is the May 1, 2026 retail rollout: 10-plus stores opened simultaneously across five cities/venues (Xuzhou Fangte theme park, Shenzhen Longgang Xinghe COCO Park, Shenzhen K11, Guilin Elephant Trunk Hill, and Yangshuo West Street), drawing a claimed 100,000-plus opening-day visitors, with a flagship-cohort figure of 1,000-plus orders per robot per day during the holiday peak. The adoption/deployment funnel (FU002) makes explicit how sharply the evidence narrows: from a claimed scale of roughly 1,000 enterprises/projects, down to 100 independently reported paying customers, down to 9 named accounts and operators this chapter could actually identify, down to just 1 deployment independently corroborated by a source with no financial or ownership tie to DexForce.[CU001, CU002, CU003, CU012, CU013, CU015]

6.3 Named Customer Proof

Named customer proof is thin relative to the scale claims above, but it is real and spans both industrial and retail tracks. Embodied Global's Series B coverage names Midea, CRRC, and Wuling as industrial customers running DexForce's models on production robotic arms for bin-picking, assembly, and insertion, and separately reports a founder quote that Lens Technology was a customer before becoming a Series B investor — a related-party detail that limits how independent that reference can be treated. Pandaily's reader-view fetch of a funding article separately names Huawei and Midea as customers, but the underlying Huawei-titled article returned only an empty/broken extraction in this run, so the Huawei detail could not be independently verified and is flagged as an open question. At the May 2026 Shenzhen Cultural Expo, DexForce signed four named retail/cultural-tourism operators (Zhongyi, Xingkong Yinli, Shaanxi Hairong, and Guangdong Dadao) whose CCTV-covered booth demo (chess, ice-cream service, and blind-box robots) is a visibility proxy, not proof of paying production use. The single strongest piece of arm's-length proof is Aigazine's independent report of a W1 Pro Gen2 robot operating unsupervised inside a real Shenzhen community convenience store, handling meal prep, microwave duties, and customer delivery — a non-DexForce, non-investor outlet corroborating a live production deployment. The customer proof matrix (FU003) scores each of these accounts on evidence quality, outcome specificity, retention visibility, and production maturity side by side.[CU006, CU007, CU019, CU024, CU026, CU027]

6.4 Retention, Durability & Satisfaction Gaps

No source fetched this run discloses net revenue retention, gross revenue retention, logo churn, contract renewal rates, or a customer satisfaction/NPS score for DexForce at either the industrial or retail segment level; each of these is recorded as an explicit null with a concrete diligence ask in the retention table (TU004) rather than silently omitted. What can be documented is unit economics for the May 2026 retail launch week: a per-store holiday revenue of RMB100,000-plus, a 4.6x uplift versus a traditional (non-robotic) store, a 99%-plus store-entry-to-purchase conversion rate, a payback period DexForce describes as "as fast as 6 months" (explicitly caveated as category-dependent), a general 30%-40% foot-traffic lift around an average store versus a much larger 200% lift measured only at the flagship Guilin Elephant Trunk Hill location, and a peak single-hour throughput of 82 orders and RMB2,050 in revenue at that same flagship store. These are launch-week, promotion-adjacent figures, not steady-state retention evidence, and the range figure (FU004) presents them as disclosed spans rather than a single blended average to avoid overstating the typical store's performance using the flagship's outlier result. A customer-support hotline (400-8086-299) is confirmed to exist in DexForce's operator manual, but ticket volume, resolution time, and complaint history are undisclosed.[CU017, CU051, CU032, CU033, CU034, CU035]

6.5 Expansion & Concentration Risk

DexForce's near-term expansion driver is the "DexBot Inside" operator/franchise model: rather than selling hardware outright, DexForce licenses its embodied-AI brain to independent operators who run the retail-service business, and the company launched a nationwide operator-recruitment call immediately after publicizing the May 2026 launch-week economics. This is capital-light for DexForce, but it concentrates execution risk in operators the company does not control, and steady-state (post-promotion) unit economics for new operators are unproven. Three further concentration risks stand out. First, DexForce's single most quotable customer reference, Lens Technology, is also a Series B investor, so that reference is not an arm's-length data point. Second, TrendForce's industry-level data shows Unitree and AgiBot alone capturing nearly 80% of China's 2026 humanoid-robot shipment volume, while Unitree (raised at a much larger scale) and UBTech (publicly listed, reporting RMB2,001 million in FY2025 revenue) both dwarf DexForce's independently reported RMB100 million-plus annualized revenue — meaning DexForce competes for a small residual share against far better-capitalized rivals. Third, essentially all of the named customer proof this chapter could assemble clusters around a single Series B funding announcement or a single trade-show signing event, rather than a broad, independently sourced customer roster, which risks overstating the breadth of durable demand relative to a company with a fully independent reference base. DexForce has also stated an intent to pursue an IPO, which would eventually force fuller customer-concentration disclosure but provides no such visibility today.[CU018, CU039, CU040, CU041, CU042, CU043]

6.6 Exhibits

7.1 Risk overview and heatmap

DexForce's risk profile spans five interlocking categories: regulatory/legal, operational/quality/security, partner/dependency, people/execution, and financial/model risk. The company is an early-stage, pre-IPO humanoid-robotics vendor operating inside a fast-moving PRC regulatory regime (a 2026 national standard system, a 2026 MIIT/SASAC special action, and China's first city-level embodied-AI law in Hangzhou), while its own commercial narrative rests heavily on self-reported, largely unaudited metrics: a single holiday-period revenue snapshot, an investor-sourced revenue figure, and a benchmark ranking on one external platform. This chapter treats DexForce's own claims skeptically where independent corroboration is absent, and draws heavily on adverse third-party sources -- Morgan Stanley's shipment-forecast note, an NDRC official's bubble-risk warning, and TrendForce's competitive-concentration data -- to stress-test the growth case. The heatmap below scores DexForce's ten highest-attention risks on likelihood, impact, and mitigation maturity; the most severe combinations of high likelihood, high impact, and low mitigation maturity concentrate in the absence of independent safety certification, the undocumented authentication posture of the joint-control API, and macro sector sentiment entirely outside DexForce's control.[CR001, CR002, CR005]

7.2 Regulatory and legal risk

China's regulatory environment for humanoid robotics tightened materially in 2026. MIIT and SASAC issued a joint notice launching a national 'real-scenario training' special action, directing provinces to select at least 20 representative scenario units and central enterprises at least 10, with a stated goal of 'ten-thousand-unit-scale' deployment capability by end of 2026 -- and explicitly calling on developers to strengthen collision detection, force-control limits, emergency braking, and black-box safety capability, implying that safety engineering for these behaviors is still industry-wide work in progress. Separately, China released its first national standard system for humanoid robotics and embodied AI (HEIS 2026), and Hangzhou enacted the first binding city-level embodied-AI robotics law, introducing coding-traceability and 'sandbox supervision' mechanisms. Against this backdrop, DexForce's own W1 user manual is notably candid: it states the robot has not passed medical-safety certification and lists at least seven categories of prohibited misuse, each with an explicit hazard rationale. No source reviewed discloses an independent industrial safety certification (CE, UL, ISO, or PRC equivalent) for the W1/W1 Pro platform, nor any DexForce-specific enforcement action or litigation -- though the latter absence itself remains an unresolved gap rather than a clean bill of health.[CR003, CR004, CR006, CR007, CR008, CR009]

7.3 Operational, quality, and security risk

DexForce's own technical documentation surfaces operational risks the company does not otherwise publicize. The Humanoid API (v0.4.0) exposes ROS2 topics and services for real-time joint-level state and position-mode control without describing any authentication, authorization, or access-control mechanism in the reviewed pages -- a meaningful gap for a system capable of directly commanding a robot's joints. The Barista Operation Manual documents a perception/grasping limitation explicitly: staff are told to load only about half the paper-cup rack to avoid the robot mis-grasping multiple cups, alongside a signed pre-opening checklist and an administrator escalation hotline for faults the operator cannot resolve. Independent coverage separately reports a W1 Pro unit operating fully autonomously and unsupervised inside a Shenzhen convenience store, preparing and delivering meals without human oversight -- a public-safety exposure with no disclosed incident-response protocol. No source reviewed discloses fleet-level uptime, MTBF, warranty/RMA rates, or a safety-incident record. DexForce's own deployment-economics figures (day-one revenue, a 4.6x lift, 99% conversion, six-month payback) are attributed specifically to the Labor Day holiday period and are not independently audited. A further specification conflict -- DexForce lists 34 DOF for the W1 Pro while RobotToday's independent supplier profile lists 'up to 40 DoF' -- remains unreconciled.[CR017, CR018, CR019, CR020, CR021, CR022]

7.4 Partner and dependency risk

DexForce's commercialization story depends on a concentrated set of external nodes it does not directly control. Its commercial-service revenue narrative rests on third-party site operators across a handful of named cities, and the company states it is actively recruiting more operators nationwide -- meaning operator churn or non-renewal would directly undercut the revenue proof-points cited in its own press releases. Its capital base shows a consistent pattern of state-linked and strategic investors recurring across rounds: Shenzhen Capital Group, the Guiyang Digital Economy Fund, and Lenovo Capital reappear from the 2024 strategic round through the mid-2025 A1/A2 round to the June 2026 Series B (alongside new entrants such as Qianhai Mother Fund and ICBC Capital), a concentration that leaves financing exposed to any shift in state investment priorities. DexForce's flagship DexWorldModel benchmark claim depends on a single external evaluation platform (WorldArena) whose methodology was not independently retrieved in this research. Participation in the MIIT/SASAC 2026 special action offers a potential validation channel, but no source confirms DexForce as a named scenario-unit participant in any province. Finally, TrendForce forecasts Unitree and AgiBot alone will capture nearly 80% of 2026 China humanoid shipments, concentrating industry volume and regulatory attention on two rivals rather than DexForce. Independent trade press corroborates this concentration in production terms: AgiBot rolled its 15,000th cumulative unit off the line by June 2026, scaling from 5,000 to 10,000 to 15,000 units within roughly three years of its founding, a manufacturing-scale pace DexForce has disclosed no comparable milestone against.[CR027, CR028, CR029, CR030, CR031, CR032]

7.5 People and execution risk

DexForce discloses relatively little about its organization beyond its founder and a general team description. The company's careers page describes a core team drawn from MIT, NTU, NUS, KAIST, Tsinghua, and Zhejiang University, with R&D staff exceeding 70% of headcount, but does not disclose total headcount, attrition, or named executives beyond founder Professor Jia Kui of CUHK-Shenzhen. This creates a key-person concentration risk typical of early-stage deep-tech ventures. Talent competition is intense industry-wide: the HEIS national-standard-drafting committee itself draws senior personnel from Unitree and AgiBot, DexForce's two largest-volume rivals, competing for the same scarce pool of senior humanoid-robotics engineers. On the financial-execution side, the only independently reported revenue figure for DexForce -- 'hundred-million-RMB-level' as of mid-2025 -- traces to a single investor quote (Lenovo Capital's investment lead, citing 100+ paying customers) repeated across two Chinese financial-media outlets covering the same funding-round press cycle, rather than to an independent audit. DexForce's stated near-term use of Series B proceeds -- world-model iteration, physics-engine upgrade, data infrastructure, and humanoid-capability completion -- implies continued reliance on R&D and field-deployment headcount growth well ahead of any disclosed revenue run-rate sufficient to self-fund that investment.[CR036, CR037, CR038, CR039, CR040, CR041]

7.6 Financial and business-model risk

DexForce has raised at least three disclosed funding rounds in roughly two years -- a 2024 strategic round led by Lenovo Capital, a mid-2025 A1/A2 round, and a June 2026 Series B at a post-money valuation exceeding RMB 10 billion -- a capital-intensive, pre-profitability trajectory typical of the sector but one that leaves the valuation resting on a self-reported, unaudited revenue base. Independent adverse sources complicate the growth case materially: Morgan Stanley forecasts China's 2026 humanoid shipments at roughly 12,000 units versus some industry projections exceeding 100,000, citing 'limited working capability' as the binding constraint, and does not expect the 100,000-unit mark to be crossed before 2030; separately, a Beijing NDRC official publicly acknowledged in November 2025 that speed and bubble risk are live concerns for the sector, while Goldman Sachs was reported to have warned of overcapacity forming without actual orders. China Daily documents an active industry-wide price war (Unitree's R1 from RMB 26,900, Noetix's Bumi at RMB 9,998, Booster's K1 from RMB 29,900) that could compress achievable average selling prices across the sector, DexForce included, though no source benchmarks DexForce's own pricing against this trend. On disclosure standards, DexForce lags comparators that are further along the public-market path: Unitree's IPO filing discloses audited-style 2025 figures (~RMB 1.7B revenue, ~RMB 600M net profit), and UBTECH's live HKEX-listed market capitalization is continuously observable, while DexForce, as a private company that has only stated IPO intent, discloses neither audited financials nor a public market valuation. A referenced HKEX filing tied to broader sector commentary also returned a 404 error during this research run, blocking independent verification of that specific document. Separately, industry-wide 2026 monetization analysis documents AI companies broadly shifting from flat/seat-based pricing toward usage-based, token-based, and hybrid pricing models to align revenue with variable compute and hardware costs; DexForce discloses no public pricing or unit-economics model for its W1/W1 Pro deployments, leaving open whether its commercial model would withstand the same cost-alignment pressure. Unitree's valuation trajectory -- a $1.7 billion Series C round in mid-2025 followed by an IPO filing targeting an implied ~RMB 42 billion valuation less than a year later -- alongside UBTECH's continuously observable HKEX market capitalization, further illustrates a pace of public/private valuation re-rating among DexForce's most direct competitors that DexForce cannot be benchmarked against.[CR043, CR044, CR045, CR046, CR047, CR048]

7.7 Mitigations, monitoring indicators, and thesis-break triggers

The single largest concentration of unresolved diligence gaps for DexForce -- absence of independent safety certification, undisclosed fleet-reliability data, unaudited revenue figures, and undisclosed headcount -- collectively represent the most material blockers to underwriting the company's commercialization and financing narrative. A set of publicly observable indicators would each measurably reduce one of these risk categories if they occur: a published safety certificate, disclosed non-holiday-period deployment economics, confirmed named-operator renewal, confirmed state-program scenario-unit allocation, a funding round led by a non-state-linked investor, or a filed IPO prospectus. Critically, DexForce's individual risk factors do not sit in isolation -- they transmit into a common set of downstream outcomes (procurement access, public-safety incident exposure, revenue durability, financing availability, and ultimately valuation), such that a single severe event, such as a reported safety incident or a sharp sector-wide sentiment shift, could compound across more than one category simultaneously. The mitigation and kill-criteria table below sets out monitorable triggers -- spanning certification status, API-security disclosure, operator renewal, headcount transparency, independent financial audit, sector shipment data, state-program confirmation, and follow-on investor composition -- that should each be tracked as a condition of maintaining or reducing confidence in the DexForce investment thesis.[CR052, CR053, CR054, CR055, CR056]

7.8 Exhibits

8.1 Investment thesis and anti-thesis

DexForce's bull case rests on four pillars: a credible, multi-institution Series B (Shenzhen Capital Group, ICBC Capital, Qianhai Mother Fund, Lens Technology, Hengjian Capital) that crossed a RMB10 billion post-money valuation in June 2026; real commercial deployment evidence, including simultaneous multi-city kiosk launches; an independently-claimed world model benchmark placement (DexWorldModel's first-place WorldArena Track 2 finish); and a differentiated "physical-AI infrastructure" positioning that independent coverage compares to Scale AI or a domain-specific Nvidia Omniverse rather than to a hardware OEM. The anti-thesis is just as concrete. EmbodiedGlobal's own reporting on the Series B explicitly cautions that DexForce's headline figures should be treated as claims until third-party benchmarks are available. One Series B strategic investor, Lens Technology, reportedly invested only after first becoming a customer -- a related-party dynamic the founder himself describes as "use it first, then invest." No source reviewed for this chapter discloses a cap table, liquidation-preference terms, customer-concentration breakdown, or independent safety certification, and no independent leaderboard corroborates the WorldArena claim. Two independent, sector-level voices -- a Beijing NDRC official warning of bubble risk across China's 150+ humanoid entrants, and Morgan Stanley's own November 2025 note calling the 100,000-unit 2026 bull case "a lot further away than you think" -- apply directly to any diligence built on DexForce's self-reported momentum. Both the thesis and the anti-thesis are evidence-supported; the open question is which one the next 12-24 months of disclosure resolves in DexForce's favor. [CV001, CV002, CV010, CV006, CV008, CV009]

8.2 Recommendation, confidence, risk rating, and valuation stance

Weighing the thesis against the anti-thesis, this chapter's recommendation is research-more -- not track, and not buy -- with medium confidence, a high risk rating, and a stretched valuation stance. The case for caution over conviction is straightforward: DexForce's post-Series-B valuation implies an estimated revenue multiple that sits well above the disclosed multiples of its two most financially transparent China humanoid peers, Unitree (an estimated ~7.5x on its pre-IPO mark and ~25x on its targeted IPO market cap) and UBTECH (an estimated ~24.5x on its current public market capitalization), even after crediting DexForce's own guided FY2026 revenue growth. A "buy" call is not supportable while nearly every scale, revenue, and benchmark figure behind that multiple remains company-sourced and explicitly flagged as unaudited by an independent outlet covering the same round. At the same time, an outright "avoid" would understate real, independently corroborated financing momentum -- six-plus independent outlets confirm the Series B terms -- and a benchmark placement claim that at least appears in independent news coverage even if not independently reproduced. Research-more reflects that middle ground: the thesis is credible enough to keep tracking closely, but not yet supported by the audited financial, cap-table, or certification evidence a "track" or "buy" call would require. The decision-relevant thresholds that would move this call toward an upgrade or a downgrade are explicit and testable, and are developed fully in the recommendation summary table and the thesis-break triggers table later in this chapter. [CV054, CV040, CV038, CV039, CV055, CV056]

8.3 Current financing context and entry discipline

DexForce's disclosed financing history spans three events: a May 2024 strategic round led by Lenovo Capital; combined A1/A2 rounds in July 2025 worth several hundred million RMB, led by Chengdu Sci-Tech Innovation Investment and Hongtai Fund, at which point Lenovo Capital's own representative credited DexForce with 100+ enterprise customers, dozens of deployed scenarios, and roughly RMB100 million in annualized revenue; and the June 2026 Series B of RMB1 billion at a post-money valuation above RMB10 billion. Independent aggregation of Chinese financial-press coverage adds a sharper data point: DexForce reports RMB100 million in H1 2026 revenue alone, and guides full-year 2026 revenue to 3-4x of full-year 2025 after "doubling every year from 2022-2024." Entry discipline at this stage means treating every one of those figures as a company-guided estimate, not an audited actual, until an IPO filing or comparable disclosure event arrives. No source reviewed reconciles a full financing ledger across all three rounds -- exact tranche sizes, instrument type (equity vs. convertible), anti-dilution terms, or cumulative lifetime proceeds are unresolved -- so any new capital committed at or near the Series B mark carries real overhang risk on both price and structure. The round's investor base itself mixes state-linked and bank-affiliated capital with policy-aligned mandates alongside private strategics, which supports the financing-momentum narrative but does not substitute for arm's-length, return-maximizing price discovery, and no source discloses any financing or valuation update since the round closed. [CV011, CV012, CV013, CV014, CV015, CV016]

8.4 Bull, base, and bear scenarios

Three scenarios frame how DexForce's valuation could evolve from its current >RMB10B Series B mark. In the bull case, FY2026 revenue lands at or above the company's own guided 3-4x FY2025 growth (an estimated RMB300-400 million or more), and a subsequent round or IPO re-rates DexForce toward Unitree- or UBTECH-style public multiples (an estimated 20-30x), implying an estimated valuation range of RMB20-35 billion -- plausible given how quickly China humanoid-sector expectations have moved (Morgan Stanley's own 2026 shipment forecast rose from roughly 14,000 to 28,000 to 50,000 units within about six months). In the base case, revenue growth continues but below the guided pace, and DexForce raises a further private round at a modest step-up before any public listing, similar in cadence to Unitree's own multi-round path toward its STAR Market filing, implying an estimated RMB13-20 billion mark -- consistent with the overall evidence picture of real momentum alongside persistent unaudited disclosure. In the bear case, the bubble-risk warnings from a Beijing NDRC official and Morgan Stanley's more cautious November 2025 note prove directionally right for DexForce specifically, the reported holiday-period kiosk economics fail to repeat, and a flat-to-down next round or valuation stagnation follows -- echoing UBTECH's own roughly 37.5% peak-to-trough public market-cap decline between December 2023 and December 2024 before its subsequent recovery. Each scenario's assumptions, valuation logic, key risks, and probability signal are detailed in the scenario table and valuation-range figure below; all ranges are explicitly estimated and inferred from peer multiples and company guidance, not derived from an independent valuation model. [CV051, CV052, CV053, CV045, CV043, CV044]

8.5 Comparable valuation set

DexForce's closest disclosed comparables are Unitree Robotics and UBTECH Robotics, the two China humanoid peers with public financial disclosure. Unitree's STAR Market IPO filing shows RMB1.7 billion in 2025 revenue and RMB600 million in adjusted net profit against a RMB12.7 billion pre-IPO valuation (an estimated ~7.5x trailing-revenue multiple) and a targeted ~RMB42 billion IPO market cap (an estimated ~25x multiple); a separate mid-2025 Series C reportedly valued Unitree at roughly RMB12 billion against roughly RMB1 billion in trailing revenue. UBTECH Robotics (HKEX: 9880) trades at a HK$54.73 billion market capitalization as of July 2, 2026 against roughly HK$2.23 billion in trailing revenue, an estimated ~24.5x multiple, though its market cap has swung by more than 60% peak-to-trough since 2023. Applying the same arithmetic to DexForce's own disclosed figures -- roughly RMB100 million trailing FY2025 revenue, or an estimated RMB300-400 million if FY2026 guidance is realized -- against its >RMB10 billion Series B mark implies a multiple of either roughly 100x or roughly 25-33x, depending entirely on which unaudited revenue base is used. AgiBot, previously a production-scale-only comparable (15,000th unit produced June 2026, independently corroborated, with a reported first-place 2025 global shipment ranking), now also has a disclosed valuation trajectory: Reuters reports a ~$2.07 billion PitchBook mark as of March 2026 and a targeted Hong Kong IPO valuation of HK$40-50 billion (~$5.1-6.4 billion) by Q3 2026, implying roughly a 2.5-3.1x step-up from private mark to IPO target. The same Reuters reporting puts Unitree's own IPO target as high as RMB50 billion (~$7 billion), moderately above the ~RMB42 billion figure cited above -- a reminder that IPO valuation targets move as the listing process progresses and should not be treated as fixed. An earlier, March 2025 report had already placed AgiBot's negotiated valuation at roughly RMB15 billion (~$2.1 billion) in a Tencent-led round, suggesting the company's private-market mark was roughly flat for about a year before the sharper re-rating implied by its 2026 IPO target. Fourier Intelligence and Robotera, which disclose nothing on their own official sites, are now valued by third-party tracker Premier Alternatives at $1.2 billion (Aug 2025, 4.85x capital efficiency) and $1.4 billion (Mar 2026, 3.52x capital efficiency) respectively; Robotera also closed a further $200 million+ financing round in April-May 2026 led by SF Group with Sequoia China, IDG Capital, HSG, and CICC Capital participating. These third-party figures round out the direct-peer set with quantified comparables but remain estimates from an external tracker and wire-service sourcing, not company-confirmed disclosure. None of these comparisons is a clean apples-to-apples match: Unitree and AgiBot are hardware-shipment businesses, UBTECH mixes education/consumer/industrial revenue lines under HKD reporting, and DexForce's own coverage frames it as closer to a physical-AI infrastructure/platform business -- a category with no disclosed public comp found in any source reviewed for this chapter. Separately, Goldman Sachs research (via Taibo Financial) estimates global humanoid robot shipments of roughly 15,000-20,000 units in 2025 with several-fold growth expected in 2026-2027, providing external corroboration for the volume assumptions embedded in these comparables' revenue-multiple math. [CV024, CV025, CV026, CV027, CV028, CV029]

8.6 Exit readiness, thesis-break triggers, and final diligence asks

DexForce's own release frames the Series B explicitly as IPO preparation, and Series B proceeds are earmarked for iterating the world-model algorithms, upgrading the physics engine, building data infrastructure, and improving humanoid capability and real-world deployment -- consistent with pre-IPO capital deployment rather than a pure growth-equity story, though no source discloses a confirmed listing venue or timeline. Exit readiness nonetheless remains unproven: no independent safety certification has been disclosed, UBTECH's own investor-relations financial page for comparison renders as a JavaScript-only shell with no extractable data, and an attempt to verify DexForce's registered-entity details via an independent China business-registry service (CNVerify) was blocked by an anti-bot challenge, leaving even basic legal-entity verification incomplete for this chapter. A separate attempt to cross-check DexForce's funding history against a startup-funding database (Fundup) was similarly blocked, and one alternate Gasgoo article URL referencing DexForce returned a 404. A parallel attempt to retrieve a specific HKEX filing-index link for an already-listed China humanoid peer returned a 404 "page relocated" notice within a single reporting cycle, underscoring how quickly even indexed regulatory filings go stale and must be re-verified rather than cited from memory. The thesis-break triggers and final diligence asks tables below translate these gaps into concrete, monitorable thresholds: an audited revenue disclosure or IPO filing would be the single most thesis-confirming event, while a down round, a customer-concentration reveal, or a documented failure of kiosk economics outside the Labor Day holiday window would each independently break the current research-more stance toward avoid. [CV004, CV005, CV019, CV020, CV021, CV032]

8.7 Exhibits