Zhiyuan Robot

1.1 Identity, product system and operating footprint

Zhiyuan Robot, also branded AgiBot, is described by a company-profile source as a Shanghai embodied-AI company designing general-purpose humanoid and industrial robots together with the software that runs them. The same source identifies Deng Taihua and Peng Zhihui as founders, February 2023 as the founding month, and Shanghai as headquarters. Its reported product scope spans full-size humanoids, smaller humanoids, wheeled industrial platforms and related software, indicating a platform strategy rather than a single-model company. Independent reporting dates the A2 release to August 2024, while a later reference dates the commercial-grade A3 release to February 2026. A third-party profile lists 663 employees, but the retained Qichacha page is a repost of company hiring material and cannot independently validate that headcount. The chapter therefore treats employee count, office count and customer count as explicit diligence gaps rather than verified scale metrics.[CO001, CO002, CO003, CO004, CO005, CO006, CO008]

1.2 Leadership, governance and key-person exposure

Public reporting identifies Deng Taihua as CEO and Peng Zhihui as a co-founder who had already built technical visibility through robotics work and engineering roles before founding the company. The available evidence is stronger on founder identity and functional leadership than on the full board, voting rights or succession plan. Dealroom says some founders and operators are hidden on its public profile, while its cap-table view labels ownership percentages as indicative. This creates a material governance diligence gap: the investor list is extensive, but the public record does not establish control rights, board composition, liquidation preferences or how dependent execution is on the two visible technical and operating leaders. Archived job postings show technical hiring breadth across robotics systems, multimodal knowledge, speech algorithms and large-model training, which supports organizational build-out but is not evidence of current staffing levels or completed governance depth.[CO005, CO006, CO007, CO008]

1.3 Funding, valuation and scale disclosure

The assigned evidence records multiple financing milestones. A 36Kr report says a Tencent-led round closed in March 2025 and that JD.com and the Shanghai Embodied Intelligence Fund participated in a subsequent round, alongside existing investors. A later report describes a prospective Hong Kong IPO valuation of HK$40-50 billion, but also says timing and terms could change; the company-profile source states that the October 2026 private round was still closing at a reported valuation of about RMB 50 billion. The profile also reports more than RMB 3 billion raised by August 2025. These figures are not equivalent to cash available, and the retained evidence does not establish audited revenue, profitability, debt, liquidation preferences or current liquidity. The public record therefore supports financing momentum, not a fully verified capitalisation table or runway assessment.[CO009, CO010, CO011, CO012]

1.4 Milestones, policy context and unresolved diligence

The evidence shows a fast progression from founding to product launch, mass production, commercial deployment and financing. 36Kr reports the first 1,000 general-purpose embodied robot units leaving production in January 2025 and describes the company's three product lines and education, factory, reception and exhibition use cases. Forbes reports 5,000 robots shipped in three months and 10,000 total shipped by March 2026. Forbes also cautions that humanoid robotics remains nascent and that premature scaling can squander momentum if products do not deliver. These figures use different verbs and source relationships, so they should not be collapsed into one audited shipment series. The Ministry of Industry and Information Technology explicitly notes sector gaps in components, operating systems, complete products, leading firms and industrial ecosystems while calling for safety governance. The company has visible scale evidence, yet audited financials, cash burn, customer concentration, incident controls and complete governance remain open questions.[CO013, CO014, CO015, CO016]

1.5 Exhibits

2.1 Market boundary and value chain

The relevant market is the sale and deployment of humanoid and embodied robots, including complete platforms, controls, software, sensors, actuators, dexterous hands, integration, and lifecycle support. It excludes general industrial robots, fixed automation, and unrelated AI software unless those products are directly incorporated into an embodied-robot deployment. This boundary matters because published estimates use different inclusions: Mordor describes a humanoids market segmented by motion, component, end-user industry, form factor, and geography, while GM Insights covers a broader global humanoid-robot market across product types, components, applications, and regions. Zhiyuan's product range spans complete robots and components, while the Gasgoo account shows its OmniHand can be compatible with robots from other manufacturers. The commercial opportunity is therefore both an OEM market and an upstream component market, but the two should not be added together without avoiding double counting.[CM001, CM002, CM003, CM004]

2.2 Sizing lenses and commercialization trajectory

Public sizing supports a large but highly uncertain opportunity. GM Insights estimates the global humanoid-robot market at USD 7.9 billion in 2025, USD 10.9 billion in 2026, and USD 192.7 billion in 2035, implying a 37.6% 2026-2035 CAGR. Mordor estimates a narrower humanoids market at USD 3.93 billion in 2026 and USD 17.80 billion in 2031, with a 35.26% CAGR. These figures are not interchangeable: they have different market boundaries, forecast horizons, and proprietary methods. The operational signal is stronger than the headline TAM: TrendForce expects Chinese output growth of up to 94% in 2026 and says the sector will focus on tangible user value in the second half of 2026. Its report also describes AgiBot production rising from 1,000 to 5,000 and then 10,000 units, but those are company-level production observations, not a market-wide revenue forecast. Zhiyuan-specific SAM and SOM remain unisolated in the public evidence.[CM005, CM006, CM007, CM008, CM009]

2.3 Buyers, adoption drivers, and constraints

The near-term buyer is more likely to be an industrial or institutional operator with a defined workflow than a general consumer. MIIT's policy interpretation identifies special environments, manufacturing, healthcare, household services, agriculture, and logistics as target application directions, with manufacturing examples including 3C and automotive. TrendForce similarly points to automotive, consumer electronics, and logistics orders as evidence of movement from trials toward demand. Budget ownership therefore sits with operations, manufacturing, logistics, facilities, or innovation teams, while engineering, safety, and IT functions influence acceptance. Adoption is constrained by integration, uptime, safety, certification, data governance, and total cost of ownership. China's 2026 standards framework covers the full value chain and lifecycle, including applications, safety, ethics, data, training, inference, and deployment. This is a market-enabling signal, but it also raises qualification requirements. Gasgoo's account of Zhiyuan's dexterous-hand spinout illustrates a strategic choice between in-house control and specialized supply; the commercial outcome of that model is not yet proven.[CM010, CM011, CM012, CM013, CM014]

2.4 Exhibits

3.1 The competitive field is already stratifying around deployment scale

Zhiyuan, also reported internationally as AgiBot, should be compared first with Unitree rather than with prototype-only developers. Forbes cited Omdia estimates of just under 5,200 Zhiyuan/AgiBot units and 4,200 Unitree units, versus roughly 150 for the first U.S. manufacturers as a cohort. TrendForce separately projects Zhiyuan and Unitree together at nearly 80% of 2026 shipments in China, making production learning and supplier access immediate advantages. Apptronik represents a different threat: lower disclosed deployment scale, but $935 million in Series A capital and factory or warehouse pilots with Mercedes-Benz, GXO Logistics, and Jabil. Figure, Agility, Tesla, Boston Dynamics, and 1X remain material entrants, while fixed automation, conventional service robots, outsourced labor, and internal integration remain status-quo substitutes where a general humanoid does not yet clear reliability or cost thresholds.[CP005, CP006, CP007, CP008, CP009]

3.2 Breadth and channel experimentation are strengths, but not proof of repeatable economics

Zhiyuan has a broader public application footprint than a single-purpose robotics vendor. Reporting describes three product lines across education, factories, reception, and exhibition guidance; the A3 adds tourism and performance deployments, while the A2-W has a reported near-100-unit industrial project. Its leasing route through Qingtian Rent may lower customer capital expenditure and extends service reach through a network spanning more than 60 Chinese cities. Supplier participation also matters: Yicai Global reports that Monkey Robot cooperates on mass production of the wheeled Expedition line at stated capacity of 200 units per month. These are credible routes to distribution and after-sales learning, but reservations, capacity, and project size are not the same as paid deployments, utilization, or recurring gross profit.[CP001, CP002, CP003, CP004, CP011, CP014]

3.3 Public comparison breaks down at price, safety, and switching cost

The assigned evidence does not provide a current Zhiyuan list price, realized lease rate, service contract, warranty, or total cost of ownership. Apptronik's only public price signal in the pool is an investor expectation of roughly $80,000 per year at future high volume, not a current offer. Even basic specifications need caution: one report states up to 10 hours of A3 operation, while a Baidu reference page states eight hours. Apptronik's use of external sensors and light curtains in designated pilot areas also shows that collaborative safety remains an implementation constraint, not a solved category feature. Without comparable uptime, intervention, certification, data-rights, and service-level evidence, neither lock-in nor multi-homing resistance can be underwritten.[CP010, CP012, CP013, CP015]

3.4 The moat is a deployment-feedback loop, and rivals are attacking every component

Zhiyuan's best-supported differentiation is the combination of shipment volume, broad product coverage, supplier coordination, leasing distribution, and early field placements. That can create a feedback loop in which more deployments improve reliability, service playbooks, and component purchasing. It is not exclusive: TrendForce attributes similar supply-chain standardization and order-driven manufacturing to the leading cohort; Unitree reports meaningful humanoid revenue and high combined segment gross margin; and Apptronik can pair deep financing with global industrial partners and Google DeepMind models. The most credible displacement risk is therefore commoditization of hardware before Zhiyuan proves workflow-specific software, safety assurance, uptime, and after-sales economics. Internal build and conventional automation remain viable where buyers value determinism over humanoid generality.[CP005, CP007, CP008, CP009, CP016]

3.5 Exhibits

4.1 Revenue model and monetization

Public descriptions support a hardware-led model spanning humanoid and wheeled robot sales, deployment support, and related software/data systems. The strongest disclosed revenue series is reported by the AgiBot profile as RMB 0.3 million in 2023, more than RMB 60 million in 2024, and RMB 1.05 billion in 2025; the source attributes the figures to Gasgoo, so they should be treated as reported rather than audited company financials. This is a sharp scale-up, but the public record does not provide a revenue-recognition policy, shipment-to-revenue bridge, mix between hardware and recurring software or services, returns, warranty reserves, or customer concentration. The Fulin Precision deployment gives a monetization anchor: an industry insider placed comparable humanoid robots at CNY400,000 to CNY800,000 per unit, while the near-100-unit agreement was described as worth tens of millions of yuan. Those are not a published Zhiyuan price list or realized gross-margin data. The underwriting implication is that reported revenue is directionally encouraging, while revenue quality remains unresolved until contracts, delivery terms, acceptance criteria, and recurring service economics are disclosed.[CI001, CI002, CI003, CI004, CI005]

4.2 Unit economics and cost structure

The operating evidence points to a cost structure dominated by hardware production, integration, deployment engineering, and ongoing operation and maintenance. At Fulin Precision, four A2-W robots reportedly handled more than 800 boxes in three hours, but the customer’s chief engineer emphasized that operation and maintenance require higher technical expertise and add to total expense. The same source estimated one robot at 0.7 of a human worker under normal operations and as much as 1.4 to 2 workers when running continuously, against a local annual labor cost of CNY80,000 per worker. These are site-level observations and estimates, not audited unit economics. The Wire China analysis adds an independent hurdle: for an 18-month payback, a robot replacing a worker earning CNY108,000 annually would need to be priced below CNY150,000 if installation and maintenance remained modest. That hurdle is materially below the CNY400,000-CNY800,000 market price range cited by Yicai, which underscores the importance of utilization, multi-shift operation, deployment reuse, service revenue, and lower manufacturing cost. Public sources do not disclose COGS, gross margin, warranty cost, installation cost, utilization, or CAC/payback.[CI006, CI007, CI008, CI009, CI010, CI011]

4.3 Capital adequacy and underwriting verdict

Zhiyuan appears to have financing access, but financing access is not the same as cash available for operations. The assigned sources report more than RMB 3 billion raised by August 2025 and a further RMB 1-1.5 billion private round nearing close in October 2026, alongside a reported approximately RMB 50 billion valuation. The 36Kr report corroborates the 2025 financing sequence, including Tencent-led March financing and participation by JD.com and the Shanghai Embodied Intelligence Fund, but does not disclose proceeds available for working capital, debt, liquidation preferences, or burn. The company profile says profitability remains undisclosed and that the company was still privately held; no assigned source supplies cash on hand, monthly burn, runway, debt obligations, or a next-round trigger. Industry evidence is also mixed: TrendForce describes rising commercialization and order-driven manufacturing, while The Wire China cautions that many deals remain proof-of-concept or policy-driven until stable operation, measurable savings, and repeat orders are demonstrated. The financial verdict is therefore conditional: shipment and revenue momentum support continued diligence, but the public economics are insufficient to underwrite margin durability or capital adequacy. The immediate diligence ask is an audited or management-prepared 2025-2026 bridge from orders and units to recognized revenue, gross profit, cash burn, and remaining runway.[CI012, CI013, CI014, CI015, CI016]

4.4 Exhibits

5.1 Product surface and workflow fit

Zhiyuan's public product surface is a family of embodied systems rather than a single software SKU. The available profile describes Yuanzheng/Expedition, Lingxi, and Genie robot families, plus GO-1 and AIDEA, while the Expedition A3 evidence specifies a full-size humanoid aimed at guidance, shopping assistance, entertainment, and light-duty grasp-and-deliver work. A3's stated 3 kg arm-end load, 2 m/s TCP speed, dual-battery design, and up to eight hours of runtime describe a commercial interaction platform, not proof of general autonomous labor. Independent factory reporting gives a more operational use case: four A2-W robots sorted more than 800 boxes in three hours at Fulin Precision. The same report says an A2-W performs at 0.7 of a human worker in normal operation and 1.4 to 2 workers when operated 24/7, figures that are site-specific estimates rather than a generalized productivity guarantee. The product implication is a portfolio that can address staged deployments, but each workflow still needs task-level validation.[CE001, CE002, CE003, CE004]

5.2 Architecture, data, and deployment model

The technology stack is unusually explicit in public materials. Zhiyuan's profile names VLA+RL, WBC, WAM, and AgiBot World, while the public repository exposes GO-1, GO-1 Air, a dataset with about 1,003,672 trajectories, model checkpoints, fine-tuning scripts, and a deployment server. The repository says the data was collected at 30 Hz for GO-1 pre-training and describes remote inference because a real robot may not have a powerful GPU. That is meaningful developer evidence of a reproducible research-to-deployment loop, but it does not establish production support levels or customer-specific uptime. Yicai adds an important deployment constraint: training relies on 95% simulated data and 5% real-world data, and unexpected situations still require continued real-world data and algorithm optimization. Initial deployments reportedly take three to four months, falling to three weeks or hours when 80% of a new environment overlaps a prior site. This supports a transfer-learning and integration-services thesis, while also identifying site variance as a principal scaling bottleneck.[CE005, CE006, CE007, CE008, CE009]

5.3 Maturity, differentiation, and control gaps

Maturity is strongest where public artifacts are concrete: an open dataset and model stack, named robot lines, factory deployments, and a reported 10,000th general-purpose embodied robot rolling off the line in March 2026. Differentiation therefore appears to combine hardware iteration, data scale, embodied-model know-how, and a growing deployment ecosystem rather than a single disclosed patent or benchmark lead. China’s policy and standards environment reinforces the direction of travel: MIIT's guidance calls for breakthroughs in the robot “brain,” “small brain,” limbs, operating systems, application platforms, and safety governance, while the 2026 standards system spans data, model training and deployment, applications, and safety and ethics. Those are ecosystem-level requirements, not evidence that Zhiyuan has completed each control. The factory account is candid that the robot stopped when approached for safety reasons and that operation and maintenance require higher technical expertise. Public evidence remains insufficient on incident rates, cybersecurity, privacy, certification, failure recovery, and service-level commitments. These gaps should be treated as underwriting blockers for high-consequence or unattended deployments.[CE010, CE011, CE012, CE013, CE014]

5.4 Exhibits

The exhibits separate disclosed product attributes from operating assumptions and unresolved diligence items. Nulls mean that the assigned public sources did not provide a verifiable value, not that the metric is zero or absent. The architecture map is a presentation of documented layers and interfaces; it does not certify reliability, safety, or commercial performance. It should therefore be read as a diligence map: documented interfaces show where evidence exists, while blank controls and unmeasured outcomes remain explicit underwriting questions for any proposed deployment.[CE003, CE007, CE014]

5.5 Exhibits

6.1 Customer segments and buying motion

Zhiyuan's visible customer motion is enterprise-led and scenario-specific rather than consumer-led. The evidence spans automotive and EV parts manufacturing, telecom, retail, entertainment, logistics, and public-facing guidance or performance use cases. Fulin Precision is described as an auto-parts supplier whose customers include Volkswagen, General Motors, BYD, NIO, and XPeng, but that does not make those automakers Zhiyuan customers. China Mobile's tender is a procurement event and therefore stronger than a logo, but the public article does not state deployment timing, acceptance criteria, or renewal. Retail and entertainment deployments show different buyer jobs: repetitive material handling, store interaction, and visitor engagement. This diversity is strategically useful, yet it also implies different integration, service, safety, and procurement requirements. Public sources do not disclose the share of revenue by vertical, buyer, geography, or channel, so apparent breadth should not be read as diversified revenue.[CU001, CU002, CU003, CU004]

6.2 Adoption trajectory and evidence quality

The adoption trajectory has several observable milestones, but the units are not interchangeable. Yicai reports a CNY124 million China Mobile tender, with CNY78 million attributed to AgiBot full-size bipedal robots; Assembly reports nearly 100 A2-W robots purchased by Fulin P.M.; and the company profile reports 100 X2 units across 100 ASD retail stores, more than 300 robots at Chimelong, and 20,000 cumulative robots produced by September 2026. Production is a company-profile metric, whereas a tender and a purchase are commercial commitments; neither proves active utilization, gross margin, service burden, or repeat orders. A Shanghai government page reprinting Jiefang Daily provides unusually concrete Fulin operating detail: a three-hour livestream described two logistics shifts and more than 800 cargo boxes per shift with almost zero errors, but the report does not provide an audited trial protocol or longer-run failure rate. The public record therefore supports increasing commercial proof, not a verified retention curve.[CU005, CU006, CU007, CU008, CU009]

6.3 Named deployments, durability, and concentration risk

Named customer evidence is strongest when it identifies a counterparty, product, use case, and transaction or operating detail. Fulin provides a named buyer and a factory workflow, while China Mobile provides a named procurer and tender value. Public profiles also name ASD, Chimelong, and other scenario partners, yet they do not disclose contract duration, renewal, uptime, paid-versus-pilot status, or customer-specific outcomes. The company profile's 100 X2 units across 100 ASD stores and more than 300 robots at Chimelong are useful scale indicators, but the denominator is a deployment count rather than a customer cohort. There is no public top-10 customer concentration, churn, NRR, GRR, contract length, or service-cost disclosure. The diligence implication is to obtain a customer-level deployment ledger and reconcile orders, acceptance, live utilization, support tickets, renewals, and expansion by account.[CU010, CU011, CU012, CU013]

6.4 Expansion path and unresolved customer diligence

The observable expansion path is land-and-expand within operational settings: a robot is first demonstrated against a job, then a larger fleet or broader scenario is described. Fulin's reported conversion from a July demonstration to nearly 100 purchased A2-W units is the clearest example, while the China Mobile tender suggests a channel or institutional procurement route rather than a single-site pilot. Zhiyuan also has multiple product families and scenario labels, which could support cross-selling, but no source quantifies attach rates, pricing, deployment lead times, integration cost, or gross retention. The evidence does not establish whether the same buyer returns for additional robots, whether deployments are financed or sold, or how much field labor is required. Until those records are provided, customer quality should be assessed as promising but unproven on durability and economics, with concentration and implementation risk treated as material open questions.[CU014, CU015, CU016]

6.5 Exhibits

7.1 Risk ranking and thesis breaks

The highest-consequence risks are regulatory/data compliance and physical safety or liability. China's Data Security Law applies to data processing in China and can reach processing outside China where it harms Chinese national security, public interests, or lawful rights; the law defines processing broadly to include collection, storage, use, transmission, provision, and disclosure (SR004). The 2026 cybersecurity amendments further emphasize AI risk monitoring and safety oversight, raise the general administrative-fine cap from RMB1 million to RMB10 million, and increase serious data-leakage penalties to RMB500,000–10 million (SR005). For a robot company collecting visual, voice, telemetry, and workplace data, the investment implication is that compliance evidence must scale with deployments, not merely with engineering capability. A thesis break is a confirmed material incident, regulator investigation, or inability to demonstrate lawful data lineage, incident response, and customer controls.[CR001, CR002, CR003, CR004]

7.2 Regulatory, legal, and cross-border exposure

The regulatory exposure is both domestic and cross-border. The Data Security Law assigns obligations to anyone processing data and requires data to be effectively protected and lawfully used (SR004). A 2026 legal guide describes China's cybersecurity framework as the Cybersecurity Law, Data Security Law, and Personal Information Protection Law, supplemented by regulations, and states that the amended Cybersecurity Law applies to overseas organizations whose activities endanger China's cybersecurity (SR006). Its practical checklist includes data mapping, classified management, technical safeguards, emergency response, breach reporting, compliance audits, and impact assessments (SR006). The 2026 China supply-chain and extraterritoriality regulations create an additional conflict for multinationals complying with foreign sanctions, export controls, forced-labor rules, or disclosure obligations, with possible trade, data-transfer, entry, and personnel consequences (SR008). Zhiyuan's overseas sales and ecosystem compatibility therefore create a monitoring requirement for jurisdiction-by-jurisdiction data flows, cloud access, components, and support obligations; no public source in this pool confirms that Zhiyuan has completed that control package.[CR001, CR003, CR005, CR006, CR007]

7.3 Operational, technical, and dependency risks

Zhiyuan's product evidence shows both commercialization momentum and execution exposure. A 42HOW report says the company operates three robot families, sells in multiple countries and regions, and put its 1,000th general-purpose embodied robot into production in January; the fetched article does not state the year for that relative date (SR001). A Baidu Baike entry reports a March 28, 2026 10,000th general-purpose robot production milestone, first-batch A3 delivery in April, and an eight-hour full-charge runtime claim for the A3 (SR003). These figures are not independently reconciled in the pool and should not be treated as proof of paid deployment, uptime, or unit economics. The same product expansion increases exposure to battery, actuator, software, sensor, field-service, and integration failures. Humanoid-law analysis identifies safety, accountability, biometric-data handling, and uncertain autonomy as structural risks, including uncertainty over whether the manufacturer, operator, or software provider is responsible after harm (SR007). Public evidence also does not identify Zhiyuan's supplier concentration, failure rates, recall history, service-level performance, or customer concentration. Those omissions are material diligence gaps rather than evidence that no incident or dependency exists.[CR008, CR009, CR010, CR011, CR012]

7.4 Mitigations, monitoring indicators, and diligence asks

The available evidence points to a control agenda rather than a verified mitigation record. The legal guidance recommends documented data mapping, emergency response, breach reporting, technical safeguards, and periodic compliance audits (SR006); the amended cybersecurity regime makes prompt, documented remediation relevant to mitigation of sanctions (SR005). For Zhiyuan, management should provide a deployment-level data inventory, retention and deletion rules, model-training provenance, cross-border transfer map, incident register, safety case, human-override design, supplier dependency map, warranty and recall policy, and customer acceptance metrics. Monthly monitoring should track reportable incidents, near misses, warranty returns, mean time to repair, production-to-paid-deployment conversion, service gross margin, top-customer share, and regulator or customs actions. Kill criteria should include a material safety event without a verified root cause, a reportable data incident without timely notice and remediation, a supplier or export-control interruption that prevents contracted delivery, or evidence that public production milestones are not translating into commercially accepted deployments. Financial verification is also weak: GetLatka displays $10.8 million of annual revenue for June 2025 while explicitly saying the source was not recorded, and separately displays a $6.4 billion valuation without a visible source in the fetched text (SR002). Those figures are neither audited economics nor evidence of cash available. Until operating, control, and financial records are supplied, the residual exposure remains high despite credible engineering and production signals.[CR003, CR006, CR010, CR013, CR014, CR015, CR016]

7.5 Exhibits

8.1 Recommendation and entry discipline

The appropriate stance is Track, with high risk and low-to-medium confidence in any precise price. GetLatka reports annual revenue of $10.8M in June 2025, valuation of $6.4B, and total raised of $520M, while a later Zhidx report says informed sources put March 2026 valuation at $2.07B and an aspirational Hong Kong IPO valuation at RMB36.3B-45.5B. These are not reconciled marks, and the GetLatka page says its revenue source was not recorded. The result is not a basis for a buy recommendation: a buyer should require a current financing instrument, fully diluted cap table, preference stack, audited revenue and gross-margin bridge, and cash/runway evidence. A meaningful entry discount is warranted until those documents show that shipments convert into repeatable contribution profit rather than primarily funded scale.[CV001, CV002, CV003, CV004]

8.2 Investment thesis and anti-thesis

The positive thesis is operational scale: Forbes reports that Agibot shipped 5,000 humanoid robots in the three months before March 30, 2026, reaching 10,000 total, and TrendForce separately reports a 10,000th general-purpose robot milestone and a rise from 1,000 units in 2025 to 5,000 and then 10,000 within three months. The market backdrop is also supportive: Mordor estimates a $3.93B humanoids market in 2026 reaching $17.80B by 2031, although those are proprietary estimates. The anti-thesis is that production and market size do not establish realized revenue, gross margin, service cost, retention, or customer concentration. A 36Kr investigation describes a heavy-asset factory, broad capital deployment, and investor pressure for commercial results as valuation rises. The thesis therefore depends on converting scale into disclosed economics, not on shipment headlines alone.[CV006, CV007, CV008, CV009, CV010]

8.3 Scenario framework and comparable context

A defensible scenario model should use shipment-to-revenue conversion, gross margin, service burden, working capital, and financing dilution as explicit drivers. The current evidence supports directional scenarios only. Bull case: repeatable industrial and commercial deployments, a mature supply chain, and a successful listing or strategic financing improve liquidity and valuation. Base case: demand grows but price discovery remains private, financing continues to fund capacity, and the investor waits for audited unit economics. Bear case: the reported mark compresses as customers delay scaled rollouts, heavy factory costs and support obligations absorb capital, or a down-round and preference stack transfer value away from common equity. TrendForce's report that Unitree and Agibot could approach 80% of Chinese shipments is a competitive benchmark, not a valuation multiple for Zhiyuan. Mordor's forecast is likewise a market size input, not company revenue. No public source here supplies a comparable company's transaction multiple that can be applied without false precision.[CV012, CV014, CV015, CV016]

8.4 Exit readiness and thesis-break triggers

Exit readiness is not established by reported IPO planning. Zhidx says the company planned a Hong Kong listing and a 15%-25% share issuance, but the article does not provide an exchange filing, prospectus, pricing, or confirmed timetable. The diligence gate should therefore be documentary: obtain the latest term sheet, cap table, liquidation preferences, debt and guarantees, monthly burn and runway, audited revenue by product and customer, gross margin after field service, backlog cancellation terms, and evidence of repeat orders. The investment thesis breaks if the current mark cannot be reconciled to revenue and cash generation, if top-customer concentration makes shipments non-repeatable, if quality or safety incidents materially increase warranty cost, or if an IPO process reveals a down-round or adverse preference stack. Re-underwrite after each financing, major production milestone, or disclosed customer cohort update; until then, preserve optionality rather than chase private-market momentum.[CV017, CV020, CV021]

8.5 Exhibits