Tianji

1.1 Identity, Headquarters, and Business Model

Tianji’s official web properties are the Chinese homepage at tianjizn.com and an English mirror at /en/, both branding the company as “Tianji Robotics.” Registry and filing sources align on the legal entity name Guangdong Tianji Intelligent System Co., Ltd., the registered address at Building 3, No. 6 Industrial West Third Road, Songshan Lake Park, Dongguan, and a 2015-05-18 incorporation date. The official about page also uses 2015 as the founding year, so the cleanest working view is that formal registration and operating-history origin broadly align in 2015, even though some later English-language finance databases say 2017. The current business is broader than humanoid arms alone. Official pages describe a product stack spanning collaborative robots, SCARA robots, vision accessories, electric grippers, and force-controlled dual-arm/humanoid robot arms under the Marvin family, all tied together by Tianji’s in-house Fusion control system. The company’s current positioning in 2026 coverage is “embodied AI infrastructure”: it sells the force-control hardware, control modules, and robot subsystems that let model-centric humanoid and embodied-AI companies execute physical tasks. That framing matters because later valuation and market analysis should treat Tianji as a component/platform supplier with legacy industrial-robot revenue roots, not as a pure full-stack humanoid OEM.[CO001, CO002, CO003, CO004, CO005, CO006]

1.2 Leadership, Governance, and Key-Person Dependence

Publicly verifiable leadership is concentrated around chairman/legal representative/manager Chen Xi. Registry sources and profile databases consistently show Chen Xi as the legal representative, chairman, and manager, while also listing him as a material shareholder. Other named governance personnel include directors An Jie, Dong Zhonglang, Huang Shaoping, and Zou Rui; supervisor Duan Kaihua; and finance head Huang Naihao. This provides enough visibility to enumerate core management and statutory governance roles, but not enough to conclude that Tianji has a mature independent-board structure. That gap matters for diligence. Public sources do not surface independent directors, board committees, or a detailed governance charter for Tianji as a private subsidiary/affiliate of a listed parent ecosystem. The sparse public leadership footprint suggests meaningful key-person dependence on Chen Xi for strategy, fundraising signaling, and customer/investor trust. At the same time, the named finance and board personnel indicate that Tianji is no longer a one-person shell; it has at least a visible statutory governance layer. The practical takeaway for later chapters is to treat Chen Xi as the central operating executive while explicitly carrying a diligence ask for post-B-round board composition, veto rights, and whether major investors gained governance rights in 2024–2026 financings.[CO011, CO012, CO013, CO014]

1.3 Ownership, Funding History, and Capitalization

Tianji’s ownership story is more nuanced than headline coverage suggests. Historically, the company was incubated inside Shenzhen Everwin Precision and for years was effectively controlled by that parent. However, Everwin’s February 2024 filing is the primary source that resets the baseline: Everwin sold 30% of Tianji and waived preemptive rights in a capital increase, reducing its stake from 78% to 40% and causing Tianji to exit Everwin’s consolidated financial statements. Subsequent 2026 media and profile sources report Everwin around 27% after later financings, which is consistent with further dilution but inconsistent with EqualOcean’s simplified description of Tianji as a wholly owned subsidiary. That discrepancy is important enough to preserve rather than smooth over. Financing history is also partly observable and partly opaque. Public sources agree on a May 2026 combined Series B and B+ round totaling RMB1 billion with GL Ventures/Hillhouse and Meituan Strategic Investment as co-leads, plus Tencent, Gaorong, Luminous/Photosynthetic Ventures, and GGV/Granite. What is less clean is valuation: EqualOcean reported an RMB8 billion post-money figure, while 36Kr, Yicai, NetEase, and EmbodiedGlobal framed the company as nearly RMB10 billion post-money. Earlier financing is only partially visible through 36Kr/PitchHub, QCC, and Baike: a March 2024 strategic financing tied to Hidden Hill / GLP Hidden Hill, and a 2025 A or A+ round involving GGV, SAIF, and CMBI, with amounts undisclosed. Because the 2025 round size is not public and exact post-B ownership is not backed by a fresh filing, later valuation work should treat exact lifetime capital raised and current dilution as unresolved diligence points rather than hard facts.[CO007, CO008, CO009, CO010, CO023, CO024]

1.4 Scale Signals, Milestones, and Adverse/Conflicting Evidence

Tianji’s scale narrative strengthened materially in 2025–2026. Official and media sources say the company moved from more than 800 customers and 10,000 robots online by end-2022 to more than 2,000 force-controlled humanoid dual-arm deliveries in four months during 2025, then to more than 10,000 units of first-quarter 2026 orders spanning 45 global humanoid robot or embodied-AI manufacturers. Product pages support the idea that this is not a vaporware story: the Marvin line has concrete seven-axis specs, published force-control parameters, and certifications. The company is therefore best described as a real hardware supplier with demonstrated delivery history, not merely a fundraising concept. The adverse side is subtler but real. Everwin’s 2024 filing showed Tianji generating only RMB52.27 million of revenue with a RMB9.99 million net loss for the first nine months of 2023, alongside parent guarantees and RMB15.6 million of financial support outstanding at the time. Registry/legal-risk sources also show routine civil litigation and hearing notices rather than a clean zero-dispute profile. Finally, some scale and identity claims conflict across sources: 2015 versus 2017 founding, 166 insured employees versus 300+ staff, and RMB8 billion versus nearly RMB10 billion valuation. None of those contradictions breaks the overall investment thesis, but they do mean this chapter should serve as the canonical place where later chapters inherit qualified facts, not simplified marketing language.[CO015, CO016, CO022, CO033, CO034, CO036]

1.5 Exhibits

2.1 Market Boundary, Included Spend, and Substitutes

Tianji is not competing across the entire industrial-robot market. Its official product and corporate pages position the company around force-controlled collaborative robots, dual-arm and humanoid manipulator modules, 3D vision, electric grippers, and related intelligent-system solutions rather than around high-speed fenced welding lines or commodity six-axis cells. The Marvin family in particular is pitched for research, commercial, quasi-industrial, dual-arm, and humanoid scenarios, with 7-axis arms, integrated joint torque sensors, and optional six-axis force sensing. That makes the relevant spend boundary broader than a single “cobot arm” SKU but narrower than the total industrial-automation market: the chapter includes collaborative arms, dual-arm manipulation cells, force-control software and sensing, mobile- manipulator adjacencies, and embodied-intelligence hardware where buyers need shared-space deployment, force sensitivity, quick changeover, or dexterous two-arm work. It excludes most heavy-duty fenced robot lines, low-cost fixed-function automation that does not require force control, and generic industrial-robot demand where payload-speed performance matters more than compliant manipulation. Status-quo substitutes therefore include traditional industrial robots, manual labor, custom hard automation, and single-arm cobots from incumbents such as Universal Robots, ABB, and KUKA. For embodied-intelligence programs, substitute spend also includes prototype humanoid platforms, mobile manipulators, and internally integrated research rigs. [CM001, CM002, CM003, CM004, CM005, CM006]

2.2 Market Sizing Using Multiple Lenses

Evidence supports several market lenses, but not a single clean point estimate for Tianji’s serviceable market. Public analyst-market-data pages disagree even on the size of the global cobot category: MarketsandMarkets sizes 2025 at USD 1.42 billion with a 2030 forecast of USD 3.38 billion, Mordor Intelligence sizes 2026 at USD 2.28 billion with a 2031 forecast of USD 5.72 billion, and Fortune Business Insights sizes 2026 at USD 2.80 billion with a 2034 forecast of USD 13.27 billion. Those differences are too large to average away; they reflect methodology, service-scope, and geography differences and should be preserved as contradictory estimates. A second, deployment-based lens is less ambiguous: MIR’s 2025 China cobot bluebook says China sold more than 30,000 cobots in 2024, up 28% year on year, while global cobot sales reached 68,000 units; the same MIR sources expect China to exceed 40,000 units in 2025 and approach 100,000 units by 2028. A third lens is broader industrial adoption: IFR records 295,000 industrial robot installations in China in 2024, equal to 54% of global demand, with electronics, automotive, and metal/machinery as the three largest sectors. A fourth lens is the embodied-intelligence adjacency: China Daily cites IDC data for 18,000 embodied-AI industrial robot shipments globally in 2025, expected to exceed 50,000 in 2026, and a State Council Development Research Center forecast of RMB 400 billion domestic embodied-AI market value by 2030. These lenses imply a large TAM and expanding adjacency, but the public record does not isolate the dual-arm force-controlled subset or Tianji’s credible SOM; that remains a diligence gap rather than a number to invent. [CM010, CM011, CM012, CM013, CM014, CM015]

2.3 Buyer, User, Payer, and Adoption Path

The buyer map is segmented less by robot morphology and more by workflow pain. Electronics and semiconductor factories want precision assembly, test handling, and compact human-shared cells; automotive and auto-component plants want screwdriving, gluing, inspection, sub-assembly, finishing, and line-side flexibility; battery and new-energy operators want glue, scanning, loading, and inspection cells; labs, education, and embodied-AI developers want force sensing, data collection, and reprogrammable dual-arm platforms. In most of these segments the day-to-day user is process or manufacturing engineering, while the formal payer sits with plant capex, automation engineering, operations, or procurement rather than with central IT. Adoption tends to move through use-case screening, proof of concept, safety/risk assessment, process tuning, and only then multi-line rollout. Public evidence from A3, ABB, UR, Flex, and Estun points to a market that is diversifying beyond classic automotive OEM cells toward electronics, logistics, life sciences, and general manufacturing. That helps Tianji because force-control and dual-arm systems are most valuable where product mix changes often, workspaces are constrained, manual dexterity still matters, or the buyer wants an embodied-AI-ready hardware stack rather than a fixed-function robot. The trade-off is sales friction: these buyers often require application engineering, integration partners, and line-qualification evidence before releasing meaningful budgets, so early pipeline breadth does not translate automatically into recurring revenue. [CM027, CM028, CM029, CM030, CM031, CM032]

2.4 Growth Drivers, Constraints, and Diligence Gaps

The main growth case is real but conditional. On the positive side, China continues to anchor global robot deployment, domestic collaborative-robot leaders are scaling quickly, and standards bodies are moving embodied intelligence from concept toward repeatable industrial deployment. ABB, UR, and Flex all emphasize stronger demand for industrial-grade cobots, AI-enabled motion, and electronics-focused flexible automation, while A3 data show cobot demand rising faster than the broader North American robot market in 2026. The constraint case is just as important. Mordor, Fortune, and MarketsandMarkets all cite labor shortages and ROI as tailwinds, but they also highlight upfront integration cost, payload-speed limitations, and brownfield compatibility issues. KUKA and Teradyne add a cyclical warning from public-company disclosures: weak economies, project delays, and customer capex restraint can abruptly slow orders in robotics and electronics. For Tianji specifically, the biggest unresolved issue is conversion quality. Public sources show attractive applications and broad category growth, but they do not disclose Tianji’s live mix of research sales versus scaled factory deployments, realized payback periods, or repeat-order depth in its target verticals. The chapter therefore preserves three contradictions rather than forcing false certainty: market values differ materially across publishers, embodied-AI enthusiasm is outrunning standardized procurement in many plants, and force-controlled dual-arm hardware may be strategically important before it is statistically separable as a standalone market. [CM042, CM043, CM044, CM045, CM046, CM047]

2.5 Exhibits

3.1 Landscape and alternative set

Tianji is not just competing against one other dual-arm robot. The retained evidence shows at least five credible alternative paths for the same buyer job. ABB YuMi is the clearest integrated dual-arm incumbent for fine assembly, but its tiny 500 g payload keeps it concentrated in small-parts work rather than broader industrial manipulation. Universal Robots, Doosan, UFACTORY, and Dobot matter even without native dual-arm form factors because many industrial tasks can still be decomposed into one arm, two coordinated generic arms, or a single arm plus fixtures, vision, and software. Franka stays relevant where research labs and physical-AI developers prioritize developer friendliness over mass manufacturing. AgiBot is not a direct substitute today for every Tianji deployment, but it is the most important likely entrant because it is building an embodied-AI stack that bundles robot bodies, datasets, teleoperation, and ecosystem capital. That means Tianji has to defend itself simultaneously against precision incumbents, ecosystem incumbents, low-cost open alternatives, and vertically integrating embodied platforms.[CP001, CP002, CP013, CP014, CP015, CP017]

3.2 Capability and pricing comparison

Tianji’s public edge is clearest on the mechanical side of embodied manipulation. Its Marvin pages emphasize 7-axis force-controlled arms, synchronized dual-arm control, EtherCAT, and collaborative-robot certifications, while outside reporting adds shipment and order scale that most research-centric peers do not publicly match. But the same evidence also shows that the market is not standing still. UFACTORY exposes very transparent xArm ecommerce pricing and an unusually open developer stack. Dobot stresses safety certifications, ecosystem plugins, and process packages that shorten deployment in mainstream industrial workflows. UR competes less on integrated dual-arm hardware than on a polished software-and-support stack, while ABB still owns the symbolic high ground for true dual-arm cobots. Public pricing remains frustratingly incomplete across Tianji, ABB, UR, Doosan, and Franka, so the strongest supported conclusion is not that Tianji is cheapest or most expensive, but that buyers can compare Tianji against a field where only a few alternatives expose clean list-price anchors and many others sell by quote.[CP003, CP004, CP005, CP008, CP009, CP010]

3.3 Distribution, trust, and switching costs

The most important competitive gap is not actuator performance; it is distribution and operational confidence. UR’s retained pages foreground software, ecosystem, partner guidance, and service, and Teradyne’s filing shows that UR sits inside a larger public-company robotics segment with a global installed base. Doosan pairs industrial applications with a public IR data room and distributor-led channel posture. ABB and Doosan both benefit from being interpreted by buyers as incumbent automation vendors rather than as a narrow embodied-AI component supplier. Tianji is moving to close that gap by using its 2026 financing to build overseas sales and technical support, but its public pages still look more product-forward than channel-forward. That matters for switching cost. Early in a project, buyers can often multi-home by testing Tianji beside UR-, Doosan-, UFACTORY-, or Dobot-based cells. Switching costs become meaningfully higher only after a customer has tuned synchronized dual-arm behaviors, force-control parameters, safety validation, and data-collection workflows around Tianji’s specific architecture.[CP008, CP010, CP012, CP017, CP019, CP020]

3.4 Moat durability, adverse evidence, and likely entrants

Tianji’s moat is real, but it is not yet unassailable. The durable part is the combination of early industrial installed base, force-control know-how, synchronized dual-arm architecture, and enough shipment volume to plausibly accumulate data that newer embodied start-ups do not have. The fragile part is that much of the underlying arm hardware market is moving toward better safety, better tooling, broader ecosystems, and more capital. ABB is already framing the category around industrial-grade performance, AI, and regulatory readiness. Teradyne’s filing shows that even the market leader is not immune from robotics revenue pressure, which is evidence that cobot hardware can commoditize and cycle with demand. Franka’s insolvency-era acquisition also shows that technical credibility alone does not guarantee durable economics. Most importantly, AgiBot is deploying capital not only into robots but into datasets, teleoperation, and developer ecosystems, which is exactly the layer that could let an embodied platform internalize arm suppliers instead of depending on them. Tianji therefore looks best positioned where customers specifically need force-controlled dual-arm execution now, but less insulated where buyers can accept generic arms or where larger embodied platforms can bundle the full stack.[CP009, CP021, CP024, CP025, CP035, CP036]

3.5 Exhibits

4.1 Revenue streams, pricing and recognition issues

Public evidence supports a mixed hardware-plus-service model, but not a clean revenue ledger. Tianji's official surfaces show standardized robot products, vertical application selling, software support, spare parts, fault repair, and training, so the economic model is broader than one-time hardware shipment alone. At the same time, there is still no public list pricing for Marvin or for the legacy industrial robot portfolio; the strongest pricing signal in retained evidence is historical and relative, not current and contractual. The parent's 2025 annual report gives a narrow disclosed anchor of CNY100 million in humanoid-robot precision-component revenue, but that figure is not the same thing as Tianji standalone revenue or recognized revenue by SKU. A Tencent-hosted feature later cited about CNY200 million of 2025 Tianji revenue across collaborative-arm and humanoid-dual-arm business, but that media figure cannot be reconciled cleanly against the parent filing, so revenue quality remains partly obscured by scope mismatch and absent subsidiary financials.[CI001, CI002, CI004, CI006, CI011, CI031]

4.2 GTM motion and sales-efficiency proxies

The GTM motion looks direct and engineering-led rather than channel-led. Official pages push vertical use cases in 3C electronics, apparel, construction machinery and automotive manufacturing, then route buyers to direct phone and email contacts, training, repair and software support. That operating posture fits enterprise robotics selling more than reseller distribution. Public sales-efficiency metrics, however, are thin: there is no disclosed CAC, payback, sales-cycle length, win rate, or backlog conversion curve. The best public proxies are customer count, installed base, repeat-order signals and current order momentum. Tianji's own pages said it had already exceeded 800 customers and 10,000 robots online by end-2022, while a historical official article said price cuts tied to scale helped repurchase behavior. Company-linked 2025-2026 coverage adds the stronger near-term signal: more than 2,000 humanoid dual-arm units delivered in four months, over 100 customers served, and over 10,000 orders on hand in Q1 2026 across 45 OEM and embodied-AI customers. Those are demand signals, not revenue-conversion proof.[CI003, CI005, CI006, CI007, CI034, CI035]

4.3 Cost structure, gross-margin drivers and working-capital load

The public picture is unmistakably capital intensive. Parent filings show Everwin still funds robotics expansion inside a much larger precision-manufacturing system, and the HKEX application proof surfaces exactly where the burden lands: borrowings rose sharply in 2025, trade receivables and inventories both expanded, and year-end current assets swung into current liabilities. Those are classic signs of a business scaling physical production faster than cash conversion. Parent 2025 operating cash flow stayed positive, but it fell sharply year over year, and Q1 2026 turned negative. Gross margin for Tianji itself is not disclosed, so the best public margin drivers are qualitative: automation rate, first-pass yield, export-heavy shipment mix, product standardization, and the balance between hardware, components, and support services. The parent's global manufacturing network and robotics-capacity buildout likely help fulfillment and export reach, but they also imply substantial tooling, facility and working-capital demands that subsidiary-level reporting does not break out.[CI008, CI009, CI010, CI012, CI013, CI014]

4.4 Public traction proxies versus private-metric gaps

Tianji has unusually strong shipment and order momentum for a private robotics company, but the public traction surface still stops short of full underwriting. Parent filings confirm 2025 humanoid-related component revenue and very large shipment counts, while multiple 2026 reports repeat the 2,000-plus delivery and 10,000-plus order milestones. That is enough to believe there is real customer pull. It is not enough to underwrite revenue quality or margin durability. Public sources still do not disclose Tianji standalone audited revenue, revenue recognition policy by stream, gross margin, churn or cancellation rate, backlog aging, average selling price, service attach rate, or customer concentration within the subsidiary. The result is a strange split: the company is visible enough to demonstrate commercial relevance, but still opaque enough that core P&L interpretation depends on management-room materials rather than public evidence.[CI011, CI012, CI034, CI035, CI037, CI038]

4.5 Capital adequacy and financing dependency

Capital support is coming from both the subsidiary and the parent, which is a strong signal of opportunity and of funding dependency. Tianji's May 2026 Series B/B+ added CNY1 billion for R&D, mass production and global sales buildout, while Everwin simultaneously pursued an HKEX listing to widen funding channels and disclosed a use-of-proceeds plan that includes capacity expansion, intelligent-manufacturing upgrades, global footprint, M&A, debt repayment and working capital. The parent also carries a large guarantee envelope for subsidiary financing. None of that proves distress, but it does show that scaling robotics is not being funded only out of free internal cash generation. The HKEX filing does say the group expects at least 12 months of working-capital sufficiency assuming IPO proceeds, banking facilities and internal cash resources. That statement is helpful at parent level; it does not replace a Tianji standalone runway disclosure. For the subsidiary itself, cash on hand, monthly burn, debt and intercompany support remain undisclosed.[CI019, CI020, CI021, CI022, CI029, CI030]

4.6 Financial verdict on revenue quality, margin path and diligence blockers

The financial verdict is mixed but not ambiguous. Revenue quality is directionally better than hype-stage robotics peers because there is parent-filed revenue, real shipment volume, and clear evidence of standardized products and repeatable industrial selling. But it is still not fully underwritable because the public record does not reconcile Tianji standalone revenue with the parent's narrower robotics line, does not disclose realized pricing or gross margin, and does not provide cash, burn or backlog-conversion data at subsidiary level. Margin path likely improves with higher automation, better first-pass yield and scale, yet the available filings also show elevated receivables, inventories and borrowing pressure that can absorb cash during fast growth. Capital intensity is therefore a first-order diligence issue, not a footnote. A serious investor should treat order growth as real, but treat valuation, runway and margin confidence as provisional until Tianji discloses subsidiary financials, backlog aging, customer concentration and robotics-specific capex allocation.[CI017, CI023, CI024, CI027, CI028, CI029]

4.7 Exhibits

5.1 Product definition in customer workflow terms

Tianji is best understood as a robot-cell supplier with an embodied-intelligence extension, not as a pure research-humanoid company. Official surfaces show a standardized catalog spanning industrial six-axis robots, collaborative robots, SCARA robots, food-grade variants, controllers, software, optional peripherals, and the Marvin dual-arm family. In customer workflow terms, the legacy stack addresses machine tending, precision assembly, visual sorting, and washdown handling; the newer Marvin line targets force-controlled bimanual manipulation where integrators or humanoid OEMs need dexterous arms, coordinated control, and data-collection hooks instead of a complete end-to-end humanoid. The concrete application pages matter because they show how Tianji sells into production cells: TR8 handling automotive subassemblies with CCD positioning, SR3/SR3D doing visual sort-and-place at 0.05 mm accuracy, and TR8 serving two CNC machines without waiting between cycles. That workflow evidence makes the product feel closer to deployable automation infrastructure than to a concept demo.[CE001, CE002, CE003, CE004, CE007, CE009]

5.2 Module / SKU map and operating architecture

The official architecture is layered. At the robot-body layer, Tianji exposes Marvin dual-arm systems, Pilot collaborative arms, TR and EVO six-axis robots, SR and SRZ/SRX SCARA products, and food-industry variants. At the control layer, TRC-Micro and SRC-Micro controllers expose memory, IO, and encoder interfaces sized for industrial cells rather than hobby deployments. At the software layer, the company publishes MotoPlus secondary-development support, MotoSim EG-VRC virtual teaching, and a menu of optional functions such as force control, soft floating, conveyor tracking, fixed TCP, palletizing, helical interpolation, and virtual fencing. Optional hardware extends the stack with teaching pendants, EtherNet/IP IO hubs, and IP54 controller protection boxes. The result is not just a list of arms; it is a modular operating system for factory automation that Tianji can repurpose into force-controlled dual-arm embodied-AI applications. The weakness is that the public software surface is still stronger for industrial cells than for Marvin-specific APIs or security controls.[CE003, CE004, CE005, CE006, CE007, CE008]

5.3 Deployment, integration and support model

Deployment evidence points to an engineering-led integration model. Tianji publishes training modules that move from safety and coordinate-system basics through tool calibration, external signals, variables, and advanced program logic. It also publishes spare-part numbers, a download center containing manuals, videos, brochures, and CAD-style assets, and software pages that describe direct sensor, network, and serial integration. The application cases reinforce that this support stack is used to tune real cells: camera-linked orientation detection for 3C parts, CCD-guided automotive subassembly workstations, and two-CNC tending by one TR8 robot. Optional features further reduce deployment friction by handling conveyor tracking, floating under external load, or palletizing through templates rather than bespoke code. Developer-signal is strongest in recruiting: current job posts explicitly ask for humanoid-arm control algorithms, system integration, SLAM, and reinforcement learning, which implies the company is still building out embodied-control capabilities instead of freezing the stack at brochure stage.[CE018, CE019, CE020, CE021, CE023, CE024]

5.4 Differentiation, manufacturing base and product maturity

Tianji’s core differentiation is the combination of an existing industrial-robot base with a force-control manipulation push. Officially, it claims 289 patents, 33 software copyrights, more than 140 R&D staff, joint labs with HKUST, SJTU and HIT, and a quality system that uses precision testing devices, cloud traceability, and Yaskawa-linked supply chains. External 2026 coverage adds the more aggressive embodied-intelligence narrative: MEMS torque sensing, one-board dual-arm coordination, teleoperation plus data-capture infrastructure, and recent scale signals such as 2,000-plus dual-arm deliveries in four months and 10,000-plus Q1 2026 orders. Those third-party claims are still partly promotional, but they are directionally consistent with the official product matrix and with recruiting evidence around control algorithms and SLAM. The upshot is that Tianji looks more mature than many dual-arm startups on manufacturing and industrialization, while still carrying disclosure risk around which Marvin-era capabilities are proven at customer scale versus still on the roadmap.[CE002, CE013, CE015, CE016, CE017, CE028]

5.5 Trust, safety, security, compliance and residual product risk

The public trust surface is mixed. On the positive side, Marvin pages explicitly list CE, ISO/TS 15066, ISO 13849-1 PLd, EN 60204-1 and EN ISO 12100 claims; optional features add virtual fencing and soft-floating behaviors intended to constrain motion or absorb external load; and the training center teaches safety as a first-class topic. On the negative side, the website does not publish downloadable certificates, test reports, product-security policies, privacy materials for teleoperation or data collection, vulnerability-disclosure channels, SBOMs, or software patch histories. Public downloads are still weighted toward legacy industrial robots and accessories, not toward Marvin-specific controller SDKs or enterprise IT controls. That does not disprove product quality, but it means product risk remains concentrated in evidence that only management can currently provide: independent compliance artifacts, failure-rate and uptime data, field-service SLAs, and a credible security/update regime for controllers and remote-operation tooling.[CE013, CE020, CE022, CE025, CE027, CE041]

5.6 Exhibits

6.1 Customer base segmentation by buyer, user, vertical, geography, and channel

Public evidence points to a hybrid B2B book rather than a single customer archetype. On the legacy industrial side, Tianji sells standardized industrial robots, control systems and integration support into factory buyers across 3C, automotive electronics, new energy, medical, home appliances, food, footwear and adjacent manufacturing categories. In those accounts the buyer and payer appear to be plant-management, automation-engineering or procurement teams, while the day-to-day users are production, quality and maintenance operators. On the newer embodied-AI side, the buyer is better described as a robot maker or embodied-AI development team purchasing Marvin force-controlled arms, controllers and integration support as an enabling subsystem rather than a finished humanoid robot. Geography is still China-first in the public record. Tianji's contact surface exposes Dongguan headquarters plus Shanghai, East China and Southwest offices, and its support stack includes direct technical service, training, spare-parts, software and fault-repair pages. That is the operating shape of an engineering-led direct-sales model, not a reseller-led catalog motion. Overseas expansion is visible but early: 2026 media coverage says Tianji plans localized North America sales and technical-support teams, while Baike says some arms have already landed in overseas universities and research labs. The practical segmentation takeaway is that Tianji is monetizing both industrial automation budgets and humanoid-development budgets, with service intensity high enough that deployment support is part of the product rather than a side offering. [CU001, CU002, CU003, CU004, CU005, CU006]

6.2 Adoption trajectory and active deployment evidence

The installed-base story starts well before the humanoid boom. Tianji's own historical materials say it had more than 300 industrial customers and 5,000 robots online by end-2020, then more than 500 customers and 5,000 delivered robots by a 2021 conference-era article, and more than 800 customers with 10,000 robots running stably online by end-2022. A later FAIR plus exhibitor profile raises the current surface claim to more than 1,000 customers and 20,000 robots online, while also describing deep strategic partnerships with top-tier humanoid robotics enterprises. Those figures are still company-linked, but together they show a persistent installed base rather than a one-cycle marketing spike. The embodied-AI trajectory is steeper and fresher. Multiple 2026 media outlets repeat the same topline: Tianji delivered more than 2,000 force-controlled humanoid dual-arm units in four months during 2025, served more than 100 customers, and entered first-quarter 2026 with more than 10,000 units of orders across 45 global humanoid robot OEMs and embodied-AI companies. That is strong proof of demand velocity, but it is still cohort-level proof. Public sources do not name most of the 45 customers, do not disclose average order size by account, and do not show what fraction of those orders were already converted into accepted production deployments, deferred backlog, or recognized revenue. The adoption curve is therefore real, but still only partly underwritten at account level. [CU001, CU002, CU008, CU009, CU022, CU023]

6.3 Named customer proof, deployment outcomes, and reference quality

The cleanest named humanoid-customer proof is Beijing Zhongke Huiling. An Aibang Robotics article says Tianji and Zhongke Huiling reached a strategic cooperation agreement in September 2025 and completed first-batch delivery of Marvin humanoid arms. The same piece adds unusually useful deployment detail: Zhongke Huiling said it selected Marvin after evaluating multiple mainstream brands and self-developed options; the delivered configuration intentionally omitted force control because the initial tasks were grasping, handling and trajectory work; and the customer cited no overheating during acceptance and trial operation. That is much stronger than a logo wall because it describes configuration, use case and operating behavior. Outside Zhongke Huiling, the evidence mix gets weaker but still informative. Tianji's own battery-manufacturing case study gives quantified industrial outcomes — 25,000 pieces per shift, 99.5% yield, RMB1.8 million annual cost savings and six-month payback — but the customer is unnamed, which limits reference quality. Tianji's Chain Expo article places the company in Apple's exhibition area and claims the underlying technology has been used in Mac Pro manufacturing, but that remains company-issued demo-adjacent proof rather than an independently confirmed Apple procurement statement. Baike also cites a 2025 strategic cooperation with Fujian Xingyun Electronics around embodied intelligence and lithium-battery / NEV scenarios, yet public evidence still looks more like ecosystem validation than a fully disclosed recurring customer contract. The result is a ladder of proof: one named humanoid buyer with concrete technical feedback, several quantified but anonymous industrial cases, and a large unnamed 45-OEM cohort whose account quality remains opaque. [CU013, CU014, CU015, CU016, CU017, CU018]

6.4 Retention durability, concentration risk, and procurement friction

Durability is where the public record is weakest. No reviewed source discloses Tianji's NRR, GRR, churn, renewal rate, contract length, backlog aging, cancellation rate, or revenue concentration by customer. The absence itself is informative: investors can see shipment momentum and support infrastructure, but cannot yet underwrite how sticky the 45-OEM cohort is, whether industrial accounts reorder at meaningful frequency, or whether humanoid orders are spread across many small design wins or a few dominant customers. A Toutiao feature even asserts that industrial cobots and humanoid dual arms each contribute roughly half of revenue, but it provides no filing-grade support and should be treated as low-confidence media characterization rather than a canonical segmentation number. Procurement friction is easier to observe than retention. Tianji's service surface requires training, software-layer integration, spare parts and post-sale fault repair; its 2021 conference article says certain controller functions are opened to customers for secondary development; and the Zhongke Huiling proof shows configuration is tailored to task need rather than sold as a one-SKU product. That combination implies a technical, customer-specific integration cycle. On the adverse side, QCC lists a 2022 sales-contract dispute and a later civil case notice, showing commercial/legal friction is not literally zero. Meanwhile, the 2024 Everwin filing shows Tianji still relied on parent guarantees and loans while producing only modest disclosed revenue in 2023, which cautions against equating broad customer-count claims with diversified cash generation. The main commercial risk is therefore not absence of customers, but opacity on which customers truly drive durable revenue and how costly each deployment is to win and support. [CU004, CU006, CU007, CU017, CU020, CU026]

6.5 Exhibits

7.1 Severity-ranked risk outlook and investment implication

The highest-severity risks are not the headline fundraising or valuation debates; they are the risks that can break conversion, margin, or insurability after Tianji has already committed to scale. First is safety and product-liability exposure. Tianji is selling force-controlled dual-arm systems into human-shared industrial and quasi-service environments, and its own product and launch materials emphasize torque sensing, compliant control, overload tolerance, and synchronized dual-arm motion. Those are strengths, but they also mean a field failure would be measured against current Chinese robot-safety standards and against rising buyer expectations for industrial-grade collaborative performance. Second is residual Everwin dependence. Everwin has diluted from control to a large shareholder position, yet public filings still show related-party transactions, historical guarantee support, shared industrial DNA, and continuing manufacturing and customer adjacency. Third is conversion quality: Tianji publicly claims 2,000 plus 2025 deliveries and 10,000 plus Q1 2026 orders, while the closest hard financial anchors remain a lossmaking 2023 subsidiary snapshot and parent-level working-capital strain. Fourth is customer, market, and geopolitical fragility. The embodied-AI market is attracting capital rapidly, incumbents are flagging aggressive pricing and project delays, and U.S. export-control tightening around advanced AI and dual-use technology raises the chance that critical inputs or overseas go-to-market motions become harder just as Tianji pushes into North America. The practical implication is that this company can justify continued diligence, but only with milestone-based underwriting: safety, order conversion, related-party governance, and support-network readiness need to clear before the 2026 valuation can be treated as investable rather than aspirational.[CR001, CR003, CR006, CR007, CR008, CR016]

7.2 Regulatory, legal, and safety risk

Tianji’s regulatory and legal perimeter is broader than the usual private-robotics startup narrative. Registry and filing sources show a scope that covers robot development and manufacturing, industrial-system services, second-class medical-device related products, and import-export activities subject to approval. That matters because the main public risk is not a known enforcement action but the mismatch between increasingly safety-regulated deployments and still-thin public disclosure on liability controls. China now has a current stack of robot safety standards covering industrial applications, collaborative robots, and general robot safety, while ABB’s 2026 industry note highlights that buyers increasingly expect industrial durability, precision, and compliance in complex collaborative tasks. Tianji’s own materials lean into those same expectations, advertising full-axis torque sensing, optional six-axis force sensing, compliant control metrics, and 7x24 durability claims. If real-world reliability lags the marketed performance envelope, the downside is not only warranty cost; it is accident liability, deployment delays, and procurement disqualification. Legal hygiene is also imperfect rather than pristine. Aiqicha reports five filing cases, six hearing notices, and six litigation relationships. That is a modest baseline for a hardware company, but it is enough to require a docket review rather than a blanket “no litigation risk” statement. A separate compliance nuance is qualification risk: Tianji publicly touts high-tech and Little Giant-type honors, yet 2026 MIIT rules make those titles explicitly reviewable and revocable for false or inconsistent data. For diligence, the key ask is not whether Tianji has a marketing list of certifications or honors; it is whether the company can produce current safety-test records, quality escape history, product-liability insurance, and a clean explanation of each open legal matter.[CR001, CR002, CR006, CR007, CR008, CR009]

7.3 Operational, manufacturing, and dependency risk

Operationally, Tianji looks more credible than a prototype-only embodied-AI startup, but scale pressure is now obvious in the public record. Company and media sources say the post-2026 financing plan is to raise automation rate, first-pass yield, and localized overseas support while serving 45 global humanoid and embodied-AI customers and pushing 10,000-unit class delivery capacity. That wording is revealing: it describes capabilities that are still being scaled, not a fully mature operating system already in place. The manufacturing base is also less autonomous than the marketing suggests. Official materials explicitly reference a premium global supply chain, cloud-based traceability, shared Yaskawa supply-chain systems, and Everwin precision-processing support. Those ties can improve ramp speed and quality, but they also increase dependence on a wider ecosystem that Tianji does not fully control. Parent-level disclosures reinforce the operational risk. Everwin’s HKEX proof shows large inventories, rising inventory impairment, large receivables, and explicit risk-factor language around excess stock, delayed collections, and missed deliveries. Even if those figures are group-level rather than Tianji-only, they show the kind of working-capital burden that accompanies fast hardware scaling. The result is a classic robotics execution risk: Tianji may win orders faster than it hardens yield, service response, spare-part logistics, and field reliability. If that happens, the order book becomes a source of quality escapes and reputation damage instead of a moat.[CR003, CR004, CR005, CR018, CR019, CR020]

7.4 Partner, customer, market, and model risk

The deepest non-technical risk is that Tianji sits inside several concentrated networks at once. It remains linked to Everwin through shareholding history, disclosed related-party flows, and manufacturing heritage even after deconsolidation. Public cap-table reporting is still noisy enough that one favorable secondary article called Tianji wholly owned long after the parent filing showed dilution to 40%, which is a warning that investor decks may over-simplify governance reality. Customer concentration is also difficult to dismiss. Tianji cites 45 humanoid or embodied-AI customers and more than 1,000 industrial customers, but the only hard concentration numbers in retained sources are at Everwin level, where the top five customers represented 64.6% of 2025 revenue and the largest customer 23.8%. That is not a Tianji-only number, yet it demonstrates how exposed this ecosystem can be to a few anchor accounts. The demand environment is supportive but unstable. 2026 financing coverage says capital is moving toward real-world validation and engineering execution, while KUKA and Teradyne disclose fiercer pricing, project delays, and trade-policy headwinds across automation and robotics. Tianji therefore faces simultaneous risk of price compression, OEM insourcing, and backlog volatility if embodied-AI enthusiasm cools or customers delay production programs. A company with real technology can still lose value quickly if its key accounts pause, squeeze margins, or internalize the arm and control stack.[CR016, CR017, CR018, CR019, CR020, CR021]

7.5 Financial, governance, and thesis-break risks

Financially, the biggest problem is not a lack of demand signals; it is a lack of clean subsidiary-level conversion data. The 2024 Everwin filing shows Tianji was still lossmaking in the first nine months of 2023, and the parent’s 2026 first quarter report shows negative operating cash flow and higher embodied-intelligence investment pressure. The HKEX application proof adds the sharper risk architecture: elevated inventories, meaningful receivables, inventory impairment, and customer concentration all compound liquidity risk when hardware demand becomes volatile. Governance is improved versus the period when Everwin controlled Tianji, but not yet simple. Historical parent guarantees, current related-party flows, sparse public board detail, and executive centrality around Chen Xi all point to a business that still needs diligence on reserved matters, board composition, and conflict-management rules. Export-control and overseas-expansion risk amplifies the same uncertainty. BIS continues to tighten Chinese advanced-AI and chip controls, and the Commerce Control List framework means component classification and licensing discipline matter more as Tianji tries to enter North American supply chains. The thesis breaks if any one of four events occurs: a serious safety or recall incident; a revealed gap between orders and collectible revenue; renewed parent dependence that limits governance independence; or trade/compliance disruption that blocks critical components or customers. Until those are cleared, the correct investment posture is milestone-based rather than valuation-led.[CR011, CR012, CR013, CR016, CR018, CR025]

7.6 Exhibits

8.1 Current financing context and entry discipline

Tianji's May 2026 B and B+ financing is large enough to matter and small enough to force price discipline. Retained English and Chinese coverage converges on RMB1 billion of new money and a roughly RMB8-10 billion post-money valuation. That is credible evidence of strong investor demand and strong investor quality, but it does not by itself prove that the price is underwritten by public operating data. The same retained sources show why investors were interested: Tianji has a standardized Marvin dual-arm product family, a longer company history than many humanoid peers, more than 2,000 reported dual-arm deliveries in four months of 2025, and more than 10,000 units of Q1 2026 orders across 45 global humanoid OEM and embodied-AI accounts. Those are real commercialization signals. The problem is the structure gap. Public reports disclose the round size, the syndicate, and the use of proceeds, but not liquidation preferences, anti-dilution terms, board control, secondary mix, or subsidiary-level revenue conversion. A disciplined investor should therefore treat the round as proof of interest, not proof that today's price is safe.[CV001, CV002, CV003, CV004, CV005, CV006]

8.2 Comparable-set benchmarks: public automation, listed robotics, and private humanoid marks

The comparable set points in three different directions, which is exactly why Tianji is difficult to price. Large global automation leaders such as ABB, Rockwell, and Fanuc trade around roughly 5.6x to 7.5x sales, while Teradyne sits much higher at about 18.1x. Chinese listed automation peers are cheaper still on average: Inovance is near 3.7x, Siasun about 5.9x, Estun about 6.5x, and EFORT about 9.5x. Those bands imply Tianji would need something like RMB0.84-2.70 billion of annual revenue to support its current mark on industrial peer logic. Listed pure-play robotics names are the escape hatch for the bull case: UBTECH trades around 34x sales and Doosan Robotics around 120x, which would require far less current revenue to justify Tianji's price. But those higher bands come with their own warning label: they reflect scarcity value, strategic optionality, and market heat, not just proven economics. Private Chinese humanoid comps reinforce that heat. Galbot and AgiBot continue to attract very large rounds, and commentary sources place the top domestic valuation band well above RMB10 billion. Tianji belongs in that cohort conversation, but its public evidence is still thinner than its implied valuation multiple.[CV015, CV016, CV017, CV018, CV019, CV020]

8.3 Investment thesis, anti-thesis, and scenario ranges

The investment thesis is not imaginary. Tianji appears to have three things many humanoid companies still lack: a productized operating-subsystem offering, some proof of mass-production execution, and a capital syndicate willing to pay up for embodied-AI infrastructure. Official and retained media sources support a view that Tianji is trying to become the standardized “arms and cerebellum” supplier for the next wave of humanoid OEMs, rather than competing head-on as a full-stack general-purpose robot brand from day one. If that position converts the reported backlog into several hundred million renminbi of recognized revenue while preserving yield and automation improvements, the current round can still work. The anti-thesis is that the round price has gotten ahead of what public evidence can currently underwrite. There is still no standalone Tianji revenue bridge, no disclosed ASP, no gross margin, no cash-burn disclosure, and no public preference-stack detail. That means the investor is buying the option value of future scale more than the certainty of current economics. The scenario table below therefore matters more than the headline round: the base case stays below the current ask, the bull case can clear it, and the bear case resets hard if sector heat cools before revenue quality is proven.[CV006, CV009, CV010, CV011, CV029, CV033]

8.4 Recommendation, hold-or-exit logic, and final stance

The right call today is track, with medium confidence, high risk, and a stretched valuation stance. That may sound cautious relative to the strength of the operating signals, but the caution is price-sensitive rather than dismissive. Public evidence is good enough to say Tianji is not just a concept deck. It is not good enough to say the current entry is already attractive. At RMB8-10 billion, even a fairly standard venture goal of 2.5-3.0x gross value creation implies roughly RMB20-30 billion of exit equity value before any preference leakage. That requires Tianji to become one of the rare Chinese humanoid leaders that turns subsystem leadership into multi-year revenue scale, not merely a credible component vendor. Hold discipline should therefore be milestone-based: keep conviction only if the company converts orders into recognized revenue, shows gross-economics resilience, and broadens the customer base beyond a handful of OEMs. Exit or avoid if the next financing depends on hype rather than disclosed conversion data, or if adoption stalls while sector multiples compress.[CV039, CV042, CV043, CV044, CV045, CV046]

8.5 Exit readiness, diligence asks, and thesis-break triggers

Tianji is not yet visibly ready for a standalone public exit. The public capital-markets surface currently belongs to Everwin, whose filings show broader group disclosure and financing ambition, while Tianji itself still lacks a standalone public financial pack. That does not kill the investment case, but it does define the work remaining. The highest-value diligence asks are straightforward: prove recognized revenue and ASP from the reported backlog, disclose gross margin by product line, show backlog aging and cancellation rates, name major customers or reference accounts, and reveal the B/B+ preference stack. Those items determine whether today's valuation is an early win or an overpay. The thesis also has clear breakpoints. If orders fail to convert, if concentration turns out to be extreme, if yield gains do not show up after the raise, or if sector bubble concerns intensify faster than commercial adoption, the current mark can re-rate sharply downward.[CV011, CV012, CV034, CV047, CV049, CV050]

8.6 Exhibits