Hai Robotics

1.1 Identity, category creation, and business model

Hai Robotics should be understood first as a category-defining warehouse-automation vendor rather than as a generic mobile-robotics startup. In its February 2026 Hong Kong application proof, the company says it focuses on warehouse picking, the most labor-intensive warehouse task, and that it ranked as the world's largest autonomous case-handling robot provider in 2024 by both revenue and shipment volume, with market share above 30%. The same filing positions Hai as the inventor of integrated ACR solutions, tracing product evolution from the first HaiPick launch in 2017 to the 2025 HaiPick Climb release. TechCrunch's 2021 funding coverage and later company materials support that lineage: Hai chose a case-handling architecture that retrieves only needed totes or cartons rather than moving entire shelves, and it has kept building the product suite around dense storage, order staging, consolidation, and full-case handling. The business model is more robust than a one-off robot sale. The filing says Hai books meaningful initial project revenue when a warehouse system is delivered and deployed, then generates recurring revenue from after-sales maintenance packages, software, and technical and operational support. That matters because it makes the company a blended hardware-plus-software-plus-services provider whose long-term economics depend on repeat deployments, ongoing service attach, and channel relationships, not just initial hardware shipments. The filing also shows a deliberate mix shift toward distribution use cases and non-domestic markets, while official product pages emphasize the standardized HaiQ control layer and modular system architecture. In short, the overview chapter can treat Hai Robotics as a global ACR platform company with real deployment maturity and a service tail, even though later chapters still need to test how durable and profitable that model becomes under public-market scrutiny.[CO001, CO006, CO007, CO008, CO009, CO010]

1.2 Founders, governance, and leadership bench

The user-supplied founder background needed correction during research. The most authoritative public source in the reviewed corpus is the HKEX application proof, which states that Hai Robotics was founded in December 2016 by Chen Yuqi, Xu Shengdong, and Fang Bing. Chen serves as chairman and chief executive officer, Xu as chief technology officer, and Fang as chief operating officer. The filing biographies make the founder-market-fit story credible: Chen previously worked on optical-modulator controller R&D and became the company-facing strategist and fundraiser; Xu leads core product R&D with robotics training from ETH Zurich; Fang owns operations, supply chain, and mass production. This founder trio still matters operationally because the company adopted a weighted-voting-rights structure tied to the three founders, preserving enhanced control into the planned Hong Kong listing. Leadership breadth outside the founder trio has expanded materially as Hai pushed overseas. The application proof identifies Zeng Hongni as CFO, and 2025 official announcements show a thicker U.S. bench: Adrian Stoch became CEO Americas effective August 18, 2025 after senior roles at Target and GXO, while Hai also promoted Hunter Senn and Andrew Tolman and added U.S.-based leaders for customer support, software, and strategic partnerships. Those moves fit the company's shift toward non-domestic orders and local support obligations. The tradeoff is governance opacity. Even with the filing, public sources remain thinner on day-to-day board processes, committee practice, and the precise balance of control between founders and late-stage investors than they are on commercial milestones. The overview therefore supports the leadership bench as a strength, while still treating governance detail and ultimate post-IPO control dynamics as only partially disclosed.[CO001, CO002, CO003, CO004, CO005, CO038]

1.3 Capital formation, scale signals, and current traction

Hai Robotics is already operating at a materially larger scale than many private warehouse-automation peers, but the scale story mixes very strong operating signals with real underwriting cautions. The HKEX filing gives the cleanest current numbers: RMB807.0 million revenue in 2023, RMB1.36 billion in 2024, and RMB1.263 billion for the first nine months of 2025; gross margin rising from 16.0% to 26.3% to 28.9%; and more than 800 contracted customers globally by September 30, 2025. It also shows more than 1,300 projects delivered, 516 R&D staff representing nearly 36% of employees, and 2,394 patent applications. That package supports a genuine large-scale hardware-plus-software business, not a pilot-stage robotics vendor. Capital formation, however, is best read through multiple lenses. TechCrunch and PR Newswire document the 2021 B+ plus simultaneous C and D rounds totaling roughly $200 million, with backing from 5Y Capital, Sequoia China, Source Code Capital, Capital Today, and others. The 36Kr pre-IPO profile adds later-stage context, describing 15 rounds and cumulative financing of about RMB4.133 billion, plus a post-Series-E valuation around RMB10.9 billion. That valuation figure is informative but still lower confidence than the operating metrics because it comes from late-stage media synthesis rather than the filing itself. The filing's own quality of revenue is also double-edged: by 9M25 the top five customers represented 48.2% of revenue and the largest single customer 30.4%, showing that Hai has landed large strategic accounts but has not eliminated concentration risk. Investors can therefore credibly view Hai as an established unicorn-scale ACR company, while still demanding more clarity on the final IPO pricing, current private-to-public mark, and how repeatable large-customer wins remain across cycles.[CO012, CO013, CO014, CO015, CO016, CO017]

1.4 Global expansion, milestones, and the core overview risks

Operationally, Hai Robotics now presents as a global company headquartered in Shenzhen rather than a China-only robotics exporter. The filing says its footprint spans more than 40 countries and regions, with non-domestic order intake above 50% of total order intake in the first nine months of 2025 and non-domestic revenue approaching 40% of total revenue. Official company releases reinforce that direction: 2025 and 2026 announcements show a stream of integrator relationships and local-market moves through FORTNA, TGW Logistics, Hy-Tek, Conveyco, and expanded Americas leadership. Case-study pages also show use cases across retail, healthcare, 3PL, and grocery, including Boot Barn, St. Luke's, Avenue Shops, CEVA Logistics, and Umall. The year-end recap pushes the scale narrative even further with 29,000 robots in operation or underway, 1,200-plus customers, and 7,000 robots delivered in 2025. Those same sources also frame the overview risks. The pre-IPO filing and adverse commentary from Bamboo Works, Benzinga, Longbridge, and AsiaTechDaily all point to the same pressure points: Hai remains deeply loss-making under accounting metrics; liabilities have risen materially; customer concentration has increased; and the coming IPO shifts the company from a private narrative of category creation to a public test of margins, governance, and execution. Some official marketing claims also use broader ecosystem counts such as 1,200-plus customers, while the filing uses a narrower "over 800 contracted customers" definition, so readers should avoid flattening those figures into a single precise customer count. The right chapter-one conclusion is therefore not skepticism about whether Hai has built something real—it clearly has—but rather recognition that the next phase hinges on whether strong overseas growth, product standardization, and founder control translate into durable public-company economics.[CO008, CO009, CO025, CO026, CO027, CO028]

1.5 Exhibits

2.1 Market boundary and category definitions

The right starting point for Hai Robotics is not the entire warehouse-automation budget but the narrower slice of warehouse work where case-level picking, storage density, and operator travel elimination matter most. Public market reports treat warehouse automation broadly, folding together robotics, conveyors, sortation, software, and services. Hai's own filing draws a much tighter line: it focuses on warehouse picking, defines ACRs as robots that retrieve only the required case from a rack, and explicitly contrasts that architecture with shelf-based AMRs that move full shelves at ground level. That distinction matters because buyers comparing Hai are usually deciding among different ways to automate picking and replenishment, not whether to automate every warehouse function at once. For market sizing, broad warehouse automation is still useful as a ceiling lens: Mordor and TBRC both describe a 2026 global market in the mid-USD-30 billions and double-digit growth. But Hai's direct arena is closer to the warehousing-picking-automation wedge that CIC, via the filing, sizes at RMB 232.6 billion in 2026E, and narrower still to the ACR slice of that wedge at RMB 9.8 billion in 2026E. The definitional hierarchy therefore runs from warehouse automation to warehouse robotics to goods-to-person and finally to ACR. Chapter 2 should keep those layers separate, because calling the full warehouse-automation TAM Hai's direct market would overstate what its current product set and buying motion actually address.[CM001, CM002, CM003, CM004, CM005, CM006]

2.2 Buyer segments, workflows, and why Hai wins in its best verticals

The buying motion is vertical-specific even when the warehouse job is similar. Operations and supply-chain leaders usually own the business case because they feel labor scarcity, congestion, and service-level pain first, but IT and warehouse-systems teams shape whether a project is feasible inside an existing WMS or MES stack, and finance ultimately underwrites payback. Hai's fit is strongest where those stakeholders want dense storage, brownfield deployment, fast throughput, and case-level flexibility rather than a full greenfield conveyor rebuild. That is why the most consistent public use cases cluster in apparel, retail, 3PL, healthcare, and mixed manufacturing. The case corpus makes that pattern concrete. Boot Barn and Avenue Shops show apparel and retail value in the form of denser storage and faster shipping without extra walking. CEVA shows why 3PLs like flexible, omnichannel-ready capacity inside a constrained footprint. St. Luke's shows why healthcare buyers value accuracy, cleanliness, control, and service continuity. Manufacturing is a smaller revenue domain for Hai today, but the product set still supports it through lineside delivery, pallet-plus-each picking, and ERP/WMS/MES integration. Across those segments, Hai's advantage is not one magic robot metric; it is the combination of goods-to-person ergonomics, retrofit-friendly deployment, mixed-container handling, and orchestration software that reduces the disruption of adopting automation in an already-busy facility.[CM017, CM018, CM019, CM020, CM021, CM022]

2.3 Conservative TAM, SAM, and observed SOM lenses

A defensible market-sizing method for Hai uses multiple lenses with clear caveats instead of pretending that one generic number is precise. The broadest TAM comes from public warehouse-automation reports: roughly USD 34.2 billion to USD 36.4 billion in 2026 across Mordor and TBRC. A more relevant middle layer comes from Hai's filing, which cites CIC's warehousing-picking-automation market at RMB 232.6 billion in 2026E. A narrower product-category layer comes from the same filing's ACR estimate of RMB 9.8 billion in 2026E. Those three layers are not interchangeable, but together they show that Hai sells into a real and expanding automation market while still operating in a much smaller direct niche than a headline warehouse-automation TAM might suggest. The SAM proxies in this chapter are intentionally conservative. Rather than convert every possible sector into faux precision, the chapter uses function and ownership shares from Mordor to illustrate the order of magnitude of relevant budget pools: picking and packing as a functional proxy, 3PL ownership as a channel proxy, and retail or e-commerce ownership as a demand proxy. The SOM lens is even more constrained. Public data do not support a credible forward market-share forecast by geography, pricing model, or vertical mix. What is supportable is an observed historical anchor: Hai's 2024 revenue is roughly consistent with about one-third of CIC's 2024 ACR market size, which aligns with the filing's over-30% share claim. That is useful context, but it is not a promise that Hai will keep the same share as the category broadens and competition intensifies.[CM008, CM009, CM010, CM011, CM012, CM013]

2.4 Growth drivers, adoption constraints, and what matters for underwriting

The evidence base is strong on growth drivers. Mordor describes labor shortages, wage inflation, faster delivery expectations, and storage-density pressure as structural tailwinds. Hai's own product and partner materials reinforce the same point from the field: buyers want systems that can go live quickly, fit around existing racking, and scale without forcing a warehouse shutdown. Integrator channels such as Hy-Tek and Conveyco matter because they widen Hai's route to enterprise budgets, especially for operators that buy solutions through established system integrators rather than directly from robotics OEMs. Competitor messaging also supports the thesis that software and orchestration are increasingly central to value capture, which makes HaiQ strategically important even if Hai remains a hardware-forward company. The constraints are just as real. Market growth does not remove capex friction, integration risk, or macro cyclicality. Hai's filing repeatedly warns that customer budgets, long commercialization cycles, supplier constraints, safety requirements, and shifts in standards can slow adoption. Competitive substitutes are also improving quickly: AutoStore sells dense cube-storage G2P, Locus is pushing rapid-deployment robots-to-goods, GreyOrange emphasizes orchestration across heterogeneous fleets, and Geekplus remains strong in adjacent AMR categories. That mix means Chapter 2 should conclude with a balanced underwriting view. Hai is aligned with genuine market demand, especially in dense, retrofit-heavy, high-SKU operations, but the key diligence issue is whether it can defend margins and deployment economics as a still-small ACR wedge becomes more mainstream and more price-competitive.[CM038, CM039, CM040, CM041, CM042, CM043]

2.5 Exhibits

3.1 Competitive paradigms and the real landscape around Hai

The right competitive frame for Hai Robotics is not every warehouse-automation vendor, but every credible way a buyer can automate the same picking-heavy job. Hai's own filing and product pages place the company in an ACR and case-handling lane: robots retrieve specific cases or cartons from racks, move them to workstations, and increasingly combine storage, staging, consolidation, and full-case workflows inside one stack. That makes AutoStore a close substitute on dense goods-to-person storage, but not a perfect like-for-like peer; AutoStore's cube model optimizes bins in a proprietary grid, while Hai's public proof is strongest where operators want to retrofit existing buildings, push vertically, and keep cartons or eaches in native packaging. Geek+ broadens the field from another direction by spanning shelf-to-person, tote-to-person, pallet-to-person, and robot-arm picking. GreyOrange competes through a control layer that can sit above mixed fleets, while Locus and Quicktron widen the edge of the market toward autonomous in-aisle picking and modular multi-workflow systems. This matters because competitive pressure is architecture-specific. Hai's best public proof points are density, retrofit fit, and case or carton handling in constrained warehouses. AutoStore makes the strongest case for absolute cube density and mature software-backed throughput; Geek+ brings breadth and shelf-to-person scale; GreyOrange attacks from orchestration and heterogeneous-fleet control; Locus pushes piece-picking autonomy and in-aisle completion; Quicktron combines tote, shelf, pallet, and material handling under a unified control stack. Status-quo substitutes also remain real. Buyers can still choose manual picking, process redesign, or smaller conveyor retrofits when disruption risk or payback uncertainty is high. The competitive chapter therefore has to compare paradigms as much as company logos.[CP001, CP002, CP003, CP004, CP005, CP006]

3.2 Direct competitor profiles and where overlap becomes real

The most important direct overlap is still between Hai and other dense, picking-focused systems rather than with every warehouse robot brand. Hai's public edge comes from the specific combination of vertical reach, brownfield fit, and case-handling. The filing says HaiPick Climb can reach 15 meters in retrofit and greenfield sites, while the 2025 upgrade and Europe deployment show a concrete operating story around 45,000 totes per 1,000 square meters, 4,000 deliveries per hour, and native-carton workflows without manual decanting. That is an attractive bundle for apparel, retail, healthcare, and 3PL environments where space is expensive and facilities cannot be rebuilt from scratch. Rivals overlap with Hai in different ways. AutoStore is the densest mature substitute and is moving beyond bin retrieval through CarouselAI, VersaPort, and AI software. Geek+ looks broader than Hai on public evidence because it spans shelf, tote, pallet, intralogistics, and robot-arm picking under one modular suite, with public share claims that point to real scale in shelf-to-person. GreyOrange is less about one hero robot and more about owning the orchestration layer across people, robots, and systems. Locus Array is the clearest autonomous-picking threat because it brings robots directly to goods and performs work in the aisle, not only at a station. Quicktron matters because it blends Chinese mobile-robot economics with a more modular, multi-scenario architecture that can cover tote, shelf, pallet, and material-handling workflows. The field is therefore crowded, but crowded in different directions: Hai still looks differentiated, just not unopposed.[CP003, CP004, CP005, CP006, CP010, CP011]

3.3 Pricing visibility, channel power, and switching costs

Competitive outcome in warehouse automation is shaped as much by packaging and channels as by robot specifications. Hai's filing shows that channel partners still contribute a meaningful share of revenue and that the company has already partnered with six of the world's top ten system integrators. That is a real advantage because it expands local delivery capacity and gives Hai credibility inside large, complex projects. But it is not a monopoly mechanism. Geek+ also talks about certified local partners, AutoStore has a long-established partner ecosystem, and Quicktron is explicitly building a channel and modular-playbook for international markets. Integrators help Hai scale, yet they can also place multiple competing architectures in front of buyers. Public pricing evidence is thinner than architecture evidence. AutoStore provides the cleanest packaging signal through Pio's subscription-style access path and through its investor-facing disclosure surface. Beyond that, public sources mostly show project framing, modularity claims, and qualitative ROI language rather than realized price curves. That matters because switching cost and multi-homing depend on what the buyer actually commits to. Cube systems and dense rack systems can create deep physical lock-in. AMR overlays and vendor-agnostic orchestration layers leave more room for mixed fleets, phased rollouts, and later supplier substitution. Hai sits between those poles: its systems create meaningful physical and software switching friction once installed, but its brownfield-friendly design and channel-heavy motion still make it less absolute than a proprietary cube grid. Buyers therefore have to underwrite architecture plus ecosystem, not only robot speed.[CP008, CP013, CP015, CP018, CP025, CP031]

3.4 Moat durability and the crowding problem

The right cautionary lens is that Hai's moat looks real but specific. It is real because the company has a workflow bundle that many operators actually need: dense retrofit deployment, vertical reach, case or carton handling, and software-driven orchestration across storage, staging, and fulfillment. Public sources support that wedge well enough to matter. The problem is that most serious competitors are no longer telling a narrow single-robot story. AutoStore is adding AI-driven picking and more flexible workstations on top of its dense installed base. Geek+ combines shelf, tote, pallet, and robot-arm modules at large AMR scale. GreyOrange pushes vendor-agnostic orchestration across very large installed networks. Locus is trying to redefine the frontier around robots-to-goods and autonomous in-aisle order completion. Quicktron is moving toward one-platform multi-scenario automation with tote and pallet modules in the same facility. Market structure reinforces the caution. Mordor's 2026 view says mobile robots already hold a large share of warehouse-automation spend and that piece-picking is the fastest-growing subsegment. TBRC's major-player list, meanwhile, shows how many sizable vendors can credibly sell into the same buyer conversation even if they use different architectures. That means Hai is exposed on two fronts at once: product convergence and disclosure scrutiny. As Hai heads toward public markets, investors will compare its pricing, unit economics, and moat claims against more disclosed public peers such as AutoStore and Geek+, not just against private marketing narratives. The bottom line is that Hai's edge should be underwritten as a workflow-specific moat inside a crowded market, not as an uncontested platform monopoly.[CP029, CP030, CP031, CP032, CP033, CP034]

3.5 Exhibits

4.1 Revenue model and pricing opacity

Observed evidence is strong on architecture and weak on exact monetization depth. The HKEX application proof is explicit that Hai books a significant portion of revenue from one-time project delivery and deployment, then earns recurring revenue from maintenance packages, software, and operational and technical support once systems are live. That lets this chapter describe Hai as a blended hardware-plus-software-plus-services business rather than a pure equipment seller. Public sources also support a two-lane go-to-market motion: direct sales for strategic and KA customers, plus a meaningful channel partner contribution that has ranged from about one-quarter to two-fifths of revenue. Official partner pages reinforce that this is not a thin referral model; Hai is investing in partner training, pre-sales simulation, and joint market management. What the public record does not reveal is just as important. No retained source breaks out recurring mix, software revenue, maintenance attach, or ARR-like metrics. Official product pages explain HaiQ, HaiCharger, and modular systems, but they still do not publish list pricing or discount bands. Even Hai's RaaS blog should be treated carefully: it shows the company discussing a leasing-style opex model conceptually, not reporting that model as a current disclosed revenue line. The sober takeaway is therefore observed rather than inferred: Hai has a real service and software tail, but the public record is not detailed enough to price that tail or determine how much of gross-margin improvement comes from recurring monetization versus better project mix.[CI001, CI002, CI003, CI004, CI005, CI006]

4.2 Operating performance and unit-economics proxies

Observed operating scale is no longer speculative. The filing reports revenue rising from RMB807.0 million in 2023 to RMB1.36 billion in 2024, with RMB1.263 billion in the first nine months of 2025, while gross margin improved from 16.0% to 26.3% to 28.9%. That trend is directionally encouraging, and the filing ties it to better business mix: more non-domestic work, larger project sizes, higher service content, and stronger pricing terms, plus better manufacturing-domain project selection. But the path is still incomplete. Net losses remained heavy, and selling, R&D, and administrative expenses all stayed large in absolute RMB terms even as their percentages of revenue declined. Public unit economics therefore have to be handled as proxies, not as a complete model. The strongest observed proxies are the negative cash conversion cycle, the rising repurchase rate, order intake per customer, and segment margin differences. Customer case studies also help, but only cautiously. Boot Barn, CEVA, and St. Luke's provide credible evidence that Hai systems can improve storage density, throughput, accuracy, and even in-house control, yet those are operating outcomes, not disclosed payback calculations for Hai itself. The right underwriting stance is to separate observed from inferred: revenue scale and margin improvement are observed; CAC, payback, channel margin sharing, and true software economics remain unobserved in public sources.[CI015, CI016, CI017, CI018, CI019, CI020]

4.3 Liquidity, liabilities, and capital intensity

Observed liquidity data argues for caution rather than panic. Hai ended September 2025 with RMB757.5 million of cash and cash equivalents, then disclosed RMB1.876 billion of cash and cash equivalents on January 5, 2026 after a December 2025 financing. That is a meaningful improvement. The same filing also shows why capital intensity remains a live issue: operating cash flow was negative in every disclosed period, total liabilities rose to RMB6.60 billion by September 2025, and net current liabilities worsened to RMB3.90 billion. Customer advances and supplier terms help materially—contract liabilities rose to RMB1.1375 billion and even exceeded dispatched goods plus contract costs—but those advantages help fund fulfillment; they do not erase underlying burn or execution risk. The financing picture should likewise be separated into what is observed and what is inferred. Observed: 2024 and 9M25 both included series-share proceeds and fresh borrowings, and management says working capital is sufficient for at least 12 months from the document date. Inferred: because that sufficiency statement explicitly assumes redacted IPO proceeds, outside investors still cannot model true runway from public data alone. Historical funding context matters here only as background, not as a rehash of Chapter 1. TechCrunch, PR Newswire, and 36Kr together show Hai has already absorbed substantial private capital. The sober capital-intensity assessment is that Hai has real scale, genuine working-capital advantages, and fresh financing support, but it is still not demonstrably self-funding.[CI032, CI033, CI034, CI035, CI036, CI037]

4.4 Financial verdict and evidence gaps

The financial verdict is therefore mixed but not confused. Observed evidence supports a company with real industrial revenue, improving gross margins, and a business model that can reuse customer relationships through software, maintenance, support, and site expansion. Observed evidence also supports a still-heavy capital-intensity profile: large accounting losses, negative operating cash flow, rising liabilities, customer concentration, and continued dependence on external financing and prepayments. Adverse coverage from Bamboo Works and the Bamboo-authored Benzinga cross-post is directionally consistent with the filing on those points, while newer IPO summaries such as Pandaily and Sahm mostly restate the filing rather than closing the important diligence gaps. The practical conclusion is to avoid over-precision. Public evidence does not support exact claims about ARR, product-level gross margin, CAC, burn multiple, or runway months. Nor does it support confident assumptions about realized pricing or channel economics. What it does support is a narrower and more defensible statement: Hai is improving economically, but it remains a capital-intensive warehouse-automation company whose path to public-market quality still depends on mix, execution, and disclosure catching up with scale. That makes Chapter 4 fundamentally a chapter about both progress and missing data, not just progress.[CI047, CI048, CI055, CI056, CI057, CI058]

4.5 Exhibits

5.1 Product stack and workflow fit

Hai Robotics’ public product story is clearer when explained as a workflow stack rather than as a list of robot model numbers. The filing and official solution pages present HaiPick Systems as a modular family that covers three recurring warehouse jobs—dense storage, order staging and consolidation, and full-case handling—plus the newer HaiPick Climb architecture for higher-density climbing retrieval. System 1 is the baseline goods-to-person configuration; System 2 adds heavy-duty AMRs for bulk and palletized handling; System 3 adds fast-transit companion AMRs, chain-pick storage, and more explicit support for trays, cartons and original packaging. Around those systems sit HaiQ, multiple workstation formats, proprietary chargers, and a wider hardware catalog that includes telescopic, grappling-hook and fork-lifting ACR variants. That stack matters because it explains why Hai can credibly span retail, apparel, 3PL, healthcare and parts of manufacturing without publicly showing a completely different software or hardware stack for each vertical. The official download center lists manufacturing-oriented collateral, and the system pages mention lineside delivery and kitting, but the strongest public proof still sits in warehouse picking, case handling and retrofit-heavy fulfillment. In other words, manufacturing appears to be an extension of the same modular case-handling platform, not a separately documented product family. The public corpus is therefore strong on module visibility and weaker on how far the manufacturing extension goes in practice.[CE001, CE002, CE003, CE004, CE005, CE006]

5.2 Architecture, orchestration and performance

The strongest public technical evidence sits in the combination of orchestration plus modular hardware. HaiQ is described as the digital brain that links upstream ERP, WMS and MES systems to warehouse execution, equipment scheduling, data, simulation and algorithm layers. Public copy goes beyond generic software marketing: the WES explicitly covers inbound, outbound, stock consolidation and inventory checks; the ESS integrates robots, conveyors, pick-to-light equipment, emergency stops and guard doors; and the algorithm platform is described in terms of order allocation, task assignment, path planning and charging logic. The filing adds the most consequential scale claim, stating that HaiQ can coordinate up to 6,000 robots of different types at one site and that the software is intentionally standardized rather than custom-built from scratch for each project. Performance claims are directionally impressive but still need careful labeling. The filing and official pages support a coherent story around high density, short tote-delivery times, 12 to 15 meter vertical reach, fast throughput and strong accuracy. But most of those numbers are company or CIC claims rather than independent benchmark outputs. The right diligence posture is therefore not skepticism about whether the system works—the hardware, workflows and orchestration are real—but caution about taking every peak metric as universally realized. Public evidence is strongest on how the system works and where it should shine: dense case handling, brownfield adaptation, original-packaging flows, and mixed robot coordination. It is weaker on independently measured edge-case performance, energy efficiency and comparative failure behavior.[CE011, CE012, CE013, CE014, CE015, CE016]

5.3 Deployment, manufacturing and channel dependencies

Hai’s modularity story is partly a product moat and partly an implementation model. The official pages say Hai owns the intelligent equipment layer—robots, workstations, safety equipment and chargers—while allowing structural elements such as racking, totes, conveyors and fencing to be sourced locally. That matters for both brownfield economics and trade-friction resilience. It can reduce the amount of proprietary infrastructure that must cross borders, and it helps explain why partners such as FORTNA, Hy-Tek, Conveyco and Zion Solutions Group can package Hai inside larger integration programs. Public partner pages consistently position Hai as a configurable goods-to-person engine inside broader operational-design, WES/WCS and lifecycle-service offerings. The same evidence also shows why execution risk remains meaningful. Public engineering roles reference firmware setup, WMS and PLC integration, networking, site inspection, DM-code installation, LiDAR and depth-camera calibration, path-planning tuning and root-cause analysis. That is not the signature of a toy automation overlay; it is the signature of a real enterprise system with non-trivial commissioning and support demands. The upside is that this know-how is hard to replicate quickly. The downside is that standardization can only go so far before site-specific conditions, partner quality and customer process discipline start to matter. Hai’s product-tech moat is therefore not just robot design. It is the combination of modular hardware, orchestration software, and delivery playbooks that let the company and its integrators make heterogeneous sites behave like repeatable deployments.[CE020, CE021, CE030, CE031, CE032, CE033]

5.4 Trust, compliance and residual technical risk

Trust evidence improved materially in the retained 2025-2026 corpus. Hai officially announced IEC 62443-4-1 certification and described secure-development controls such as risk assessment, testing, vulnerability management and coordinated updates. Trade coverage also repeated company claims around TÜV SÜD validation of EU RED Article 3.3(d) compliance and alignment with ISO 27001 and TISAX. Official HaiQ copy adds more operational controls, including encrypted robot communications, automatic switching on failure, load balancing and disaster-prevention features. System pages also mention CE and NRTL-certified workstations and the HaiVest safety approach for human access during maintenance. This is enough to say that trust, cyber and safety are now part of the public product story, not an afterthought. What remains missing is equally important. The filing still warns that cyber-attacks, network overload, power loss and telecommunication failures can interrupt ACR availability, and it separately flags dependence on semiconductors, sensors and control systems. Public certification news does not yet provide certificate IDs, scope detail, incident-disclosure history, CVE handling records, or warranty economics. Likewise, no retained source gives independent third-party benchmarking on failure modes, energy use or service-level performance against named peers. The practical conclusion is balanced rather than binary: Hai now has meaningful public trust signals, but the most underwriting-relevant proof still depends on diligence-room artifacts and reference calls rather than what is visible in the open corpus.[CE037, CE038, CE039, CE040, CE041, CE044]

5.5 Exhibits

6.1 Customer base segmentation and who actually buys Hai

The filing is clear that Hai Robotics sells into a broader buyer universe than a simple list of end-user warehouse logos. As of September 30, 2025, the company said it had entered into contracts with over 800 customers globally, but that count explicitly includes both direct customers and channel partners. The same filing also reports 402 customers in the first nine months of 2025 in its order-intake table, which is better read as an in-period contracting-account count rather than a complete installed-base or site count. That distinction matters because it means the top-line customer story blends integrators, direct enterprise buyers, and end-user projects. Hai says the buyer mix spans warehouse operators, logistics service providers, retail conglomerates, and manufacturing enterprises across apparel and fashion, e-commerce and retail, F&B, 3PL, pharmaceutical, 3C electronics, and automotive. Enterprise quality is meaningful on paper: the filing says the customer base includes over 70 companies that have appeared on Fortune Global 500 lists since 2021, plus seven of the top 10 apparel and fashion companies and six of the top 10 3PL companies globally by 2024 revenue. The customer lens is also increasingly international. Non-domestic markets generated more than half of order intake in 9M25 and 39.6% of revenue, while the named case-study corpus now spans the United States, Spain, Singapore, Brazil, South Korea, and China. Partner pages from Hy-Tek, FORTNA, Conveyco, and Zion reinforce this mix by framing Hai as a channel-ready goods-to-person and ASRS option for verticals such as apparel, 3PL, healthcare, e-commerce, electronics, and spare parts. The right segmentation view, then, is not a single customer type but a stack: direct enterprise accounts, integrator-led programs, and channel-assisted international deployments.[CU001, CU002, CU003, CU004, CU005, CU006]

6.2 Named deployment proof: where production maturity is strongest

The strongest adoption evidence is not the headline customer count but the depth of the named production examples. Several official case pages now show enough operational detail to move beyond logo marketing. St. Luke’s and Equipment Depot provide one of the cleaner proofs because the official Hai page and an external partner announcement both describe the same healthcare insourcing project, including 28 robots, a 14,000 square foot automated footprint, and the strategic objective of bringing distribution in-house from 3PL dependence. JD Logistics is also stronger than a typical marketing case because Hai’s own page and a Honeywell-hosted PDF both describe a live California deployment with 60 ACRs, 42,028 tote locations, 100,000-plus managed SKUs, measurable throughput gains, and Black Friday flex without extra temporary labor. RD Saúde, Zuellig Pharma, Mettler Toledo, Maersk Logistics, and Scalpers add depth across pharmaceutical retail, clinical-trial logistics, electronics distribution, fashion 3PL, and omnichannel apparel. These examples show real robot counts, storage-location counts, throughput claims, and sometimes phased expansion. At the same time, proof quality is uneven. Boot Barn, CEVA, and Avenue Shops provide attractive outcomes, but the public record there remains mostly company-written. Umall and Li-Ning are more directional still: they support real usage, but not with the same level of quantified evidence. The balanced conclusion is that Hai has enough named production references to show the product is used in live operations across multiple segments, while the precision of the outcome evidence still varies materially by customer.[CU016, CU017, CU018, CU019, CU020, CU021]

6.3 Repeat usage, expansion signals, and the limits of public retention evidence

Durability evidence exists, but it is mostly proxy evidence rather than a clean recurring-revenue or renewal-cohort disclosure set. The strongest disclosed metric is the filing’s customer repurchase rate, which improved from 68% in 2023 to 80% in 2024. That is directionally encouraging, especially when combined with case-study evidence of phased or repeat expansion. Boot Barn reportedly expanded three times in one year, Scalpers is explicitly structured as a two-phase rollout from 12 robots and 20,000 active locations to 26 robots and 100,000 planned locations, and JD Logistics’ Honeywell case shows the same installed system flexing into peak season rather than behaving like a one-off pilot. These are real land-and-expand signals. But the caveat is important: the filing defines repurchase rate using direct contracting customers and channel partners, not a pure set of end-user sites or a SaaS-like cohort definition. Public materials do not disclose gross revenue retention, net revenue retention, churn, average contract length, or renewal rates. Nor do they separate how much repeat activity comes from the same warehouse, multi-site rollouts, new project phases, or partner-resold follow-on orders. Partner pages nevertheless matter here because they show the surfaces through which repeat demand can compound. Hy-Tek, FORTNA, Conveyco, and Zion all market Hai as a modular system that can be installed in brownfield facilities and expanded over time, which is consistent with the broader operational story. The result is a credible but incomplete durability picture: repeat behavior is visible, yet public evidence is still far better at proving expansion potential than at proving cohort retention quality.[CU012, CU013, CU016, CU022, CU023, CU029]

6.4 Concentration risk and how much confidence the public record really supports

The hardest customer question is not whether Hai has real customers; it is whether the quality of those relationships offsets concentration and disclosure gaps. On that score the filing is sober. Revenue from the five largest customers rose from 32.1% in 2023 to 36.7% in 2024 and 48.2% in 9M25, while the single largest customer rose to 30.4% of revenue in 9M25. Key-account concentration also climbed, with KA customers contributing 75.8% of order intake in 9M25. That does not negate the operating momentum, but it does mean public durability evidence must be weighed against meaningful dependency risk. Bamboo Works makes the same point from an adverse angle, flagging concentration as a real IPO vulnerability rather than a footnote. Another limit is evidentiary. The public named-proof set is now much broader than the five case pages used in earlier chapters because Hai’s cases index and download center surface a larger menu of customers and case assets. Even so, open-source proof still names only a thin subset of the 800-plus contracted customers in the filing or the broader 1,200-plus ecosystem claim used in Hai’s year-end marketing. Most of the visible proof is company-hosted or partner-hosted, not customer-authored disclosure. That means confidence should be highest in the existence of real deployments and lowest in the universality of the outcome claims. Investors can therefore underwrite a genuine installed base, real international reach, and some repeat expansion, but they should not overstate what public evidence proves about long-term retention durability or concentration resilience.[CU010, CU011, CU014, CU037, CU040, CU041]

6.5 Exhibits

7.1 Governance, regulatory, and legal risk

The cleanest governance risk in Hai’s case is not hypothetical; it is written into the filing. The company plans to list with weighted voting rights, and the application proof explicitly warns that the WVR beneficiaries may have interests that are not fully aligned with those of shareholders as a whole. The filing’s own risk-factor language goes further, stating that concentrated voting power could delay or prevent change-of-control outcomes and depress the ordinary-share price. That matters because this is not a founder-control structure attached to a mature public operating history; it is being introduced while the company is still in application-proof stage, before public-market discipline has been tested in live reporting cycles. HKEX Chapter 8A provides real safeguards, including a 10:1 voting cap, minimum voting rights for non-WVR holders, and cessation triggers if a beneficiary leaves the board or transfers the relevant stake. But those rules reduce the tail risk; they do not remove the practical reality that public investors could be locked into management’s strategic choices for an extended period. Legal and regulatory risk is similarly mixed. Hai’s website privacy policy shows that the company collects broad categories of user and marketing data and may transfer personal data across jurisdictions, including China and other countries, which creates real compliance surface area for a company selling connected systems globally. Its terms of use are also defensive: content is provided as-is, warranties are broadly disclaimed, and technical inspection behaviors such as reverse engineering and vulnerability scanning are restricted. On cybersecurity regulation, Hai has surfaced constructive signals, including RED Article 3.3(d) compliance messaging and IEC 62443-4-1 certification. Those are positives. But the evidence still stops at public announcements and high-level regulatory explainers rather than transparent certificate scope, surveillance cadence, or incident history. The right judgment is therefore a ranked one: WVR governance risk is observed and high; privacy and cyber-compliance obligations are observed and medium-to-high; undisclosed litigation exposure remains an explicit evidence gap rather than a cleared issue.[CR001, CR002, CR003, CR004, CR005, CR006]

7.2 Operational, support, and partner-dependency risk

The filing is unusually direct about operational fragility for an automation business. Hai says its technology infrastructure may suffer system failures, interruptions, inadequacy, security breaches, cyber-attacks, network overload, telecommunication failures, and power loss, and it explicitly links those events to lost availability, customer dissatisfaction, lower future revenue, and regulatory scrutiny. This is important because the company is not presenting a simple piece of warehouse hardware; it is presenting a live control stack that depends on robots, control systems, software, and connected networks operating together under real-world uptime expectations. The company’s own IEC 62443 explainer reinforces that point by describing warehouse robots and control systems as increasingly integrated with enterprise platforms, cloud systems, and partner networks. The cyber story therefore cuts both ways: visible compliance work and secure-development claims are real mitigants, but the integration surface is also wide, and no public incident ledger or uptime SLA is visible. Execution complexity compounds that technical risk. Hai’s current field and engineering roles require firmware work, WMS and PLC integration, acceptance testing, site inspection, network setup, safety planning, LiDAR and camera calibration, and root-cause analysis. That is exactly the kind of delivery profile that creates moat through know-how while also creating a scaling risk if too few experienced people have to support too many live sites. Partner dependence is the second amplifier. The filing says channel partners still account for a meaningful share of revenue and are essential in non-domestic markets, where they often own initial after-sales support, installation coordination, repairs, and claims handling. The filing also warns that partners may fail to comply with laws or agreements, may prioritize competitors, and may damage Hai through misconduct. Official partnerships with TGW and Dematic show real commercial leverage and contractual mitigation—training, quality, spare-parts, warranty, and liability frameworks are all visible—but they also confirm the core dependency: overseas scale is being achieved through ecosystems Hai does not fully control.[CR012, CR013, CR014, CR015, CR017, CR018]

7.3 Customer, financial, and competitive risk

The financially material risk stack is dominated by concentration and liquidity rather than by a lack of demand. The filing shows the five largest customers rising from 32.1% of revenue in 2023 to 48.2% in 9M25, with the single largest customer reaching 30.4% of revenue and KA customers contributing 75.8% of order intake in the same period. Those are not soft warning lights. They mean one customer relationship, one project timetable, or one procurement reset can move group-level outcomes materially. At the same time, Hai remains loss-making. The filing records large losses in 2023, 2024, and 9M25, rising contract liabilities tied to delivery obligations, and net current liabilities of nearly RMB3.9 billion by September 2025. Bamboo Works, Benzinga, and Longbridge all highlight the same balance-sheet and cash-flow pressure from an adverse angle. The public evidence therefore supports a specific financial risk ranking: concentration and liquidity are observed high risks; losses and delivery-obligation risk are observed medium-to-high risks; exact runway and service-margin downside remain under-disclosed. Competitive pressure is also more observed than speculative. AsiaTechDaily argues that Hai is moving into a phase where pricing, deployment speed, and customer acquisition matter more as the market crowds. That stance is consistent with the broader evidence set: warehouse automation is large enough to keep attracting credible competitors, and buyers can increasingly compare case-handling systems against broader AS/RS and AMR alternatives, not just against another ACR vendor. Hai’s own overseas growth intensifies this issue because non-domestic orders now make up more than half of order intake, which improves market opportunity but also forces the company to defend price, support quality, and local compliance in more jurisdictions. For valuation, that means investors should not underwrite Hai as a clean software-like growth story. A more realistic lens is that the company may deserve credit for category leadership and improving gross margin, but the public multiple ceiling should stay constrained until concentration eases, operating cash generation improves, and partner-led expansion proves it can scale without leaking margin or quality.[CR033, CR034, CR035, CR036, CR037, CR038]

7.4 Mitigations, kill criteria, and open evidence gaps

Hai is not ignoring these risks. The public record shows several concrete mitigations: HKEX’s WVR rulebook imposes structural guardrails; the company has surfaced RED and IEC cyber-compliance work; partner frameworks with Dematic and TGW explicitly reference training, warranties, spare parts, and liability; and Hai has added 2026 U.S. leadership for customer support, software, and partner development. Those are meaningful signals because they tie directly to the most important failure modes. But the mitigation maturity is uneven. None of the visible public sources disclose customer-facing SLA economics, incident rates, supplier concentration, warranty claim frequency, or a litigation schedule. In other words, the company is clearly building mitigation scaffolding, yet public investors still cannot tell whether the scaffolding is overbuilt, underbuilt, or economically efficient. That is why the most useful kill criteria in this chapter are concrete and monitorable rather than rhetorical. If the next disclosed period keeps the largest-customer share near current extremes, if operating cash generation does not turn convincingly positive, if support remains dependent on opaque partner handoffs without visible SLA evidence, or if cyber and regulatory assurance remains limited to marketing-level certification statements, the risk case should harden rather than soften. Conversely, the thesis improves if Hai demonstrates concentration relief, cleaner liquidity, better disclosure on support and warranty economics, and evidence that WVR governance works with restraint rather than simply because rules require it. The chapter’s final judgment is therefore disciplined: many of Hai’s top risks are already observed and rankable, several mitigations are visible, and the remaining unknowns are specific enough to turn into a diligence checklist rather than generic unease.[CR004, CR005, CR010, CR011, CR021, CR027]

7.5 Exhibits

8.1 Recommendation and valuation framing

Hai Robotics is easier to respect than to price. The business is clearly real: the HKEX application proof shows industrial-scale revenue, materially better gross margin, more than 800 contracted customers, and visible repeat-deployment behavior. Official product and customer sources also show that Hai is not simply shipping robots into warehouses; it has a software layer, customer-specific deployment proof, and enough partner infrastructure to support international expansion. Those are the ingredients of a company worth tracking, not dismissing. They are also why this chapter does not end in “avoid.” But the evidence is still not clean enough for a buy-style recommendation. The filing remains a draft application proof with no final public price range, the late-stage private valuation signal comes from media coverage rather than an issued term sheet, and the company is still loss-making, cash-burning, and highly concentrated into large customers. That combination matters because public markets do not only value industrial progress; they value disclosure quality, entry discipline, and downside protection. The decisive call from the current record is therefore track / research-more with medium confidence, high risk, and a stretched valuation stance if one starts from the reported late-stage private mark. The right question is not whether Hai has built something meaningful. It clearly has. The right question is whether the price can be underwritten cleanly from public evidence today. The answer is still no.[CV001, CV004, CV005, CV007, CV008, CV010]

8.2 Private mark versus public-market discipline

The clearest publicly visible valuation signal for Hai is the 36Kr account of a roughly RMB10.9 billion post-Series-E mark after about RMB4.133 billion of cumulative financing. That signal is useful, but it is not self-validating. When paired with Hai’s disclosed 2024 revenue, it implies about 8.0x trailing sales before any adjustment for liquidation preferences, dilution, or IPO discount. That is not an absurd number for an automation company with real growth and category leadership, but it also is not obviously cheap when the filing simultaneously shows large losses, negative operating cash flow, heavy liabilities, and worsening customer concentration. Private narrative and public valuation discipline are not the same thing. Public comps sharpen that point. AutoStore screens around 7.4x trailing sales with a disclosed reporting cadence, live public price discovery, and mature product packaging. Symbotic screens far richer at roughly 12.7x trailing sales and 11.9x EV/sales, but it also has much greater scale and public-market disclosure. Geekplus matters less as a normalized multiple and more as a market-read: it is a live Hong Kong warehouse-robotics listing with WVR and real IPO proceeds, suggesting investor appetite exists for the category. The right conclusion is not that Hai deserves one of those numbers by analogy. It is that Hai currently sits between a credible private late-stage signal and a harder public comp set, with enough quality to stay on the list but not enough cleanliness to deserve a premium entry by default.[CV002, CV016, CV017, CV018, CV019, CV020]

8.3 Scenario range and decision triggers

The right way to frame Hai’s value is through scenarios, not through fake precision. A bear case starts with the simple fact that the offering is still conditional and that public investors may not be willing to honor a late-stage private mark when losses, concentration, and disclosure gaps remain so visible. In that case, Hai would likely trade at a discount to the private signal, especially if cash generation or concentration worsens. The base case is more balanced: Hai keeps growing, margins keep improving, overseas mix stays supportive, and investors give some credit for software, services, and repeat deployments, but they still refuse to pay a premium for opacity. The bull case is possible, but it requires more than continuing the current narrative. It needs evidence that Hai can translate category leadership into cleaner recurring economics and a more public-ready risk profile. That is why the ranges below should be read as disciplined decision bands, not as live underwriting advice. They tell the investor what evidence would justify paying more or less, and they show where the current late-stage private signal sits inside that map. If concentration remains extreme, if the next update does not improve cash-generation quality, or if a future price range asks investors to pay at or above the private multiple without closing the disclosure gaps, the thesis should weaken, not strengthen. Conversely, if Hai brings out a cleaner prospectus, shows continuing mix improvement, and gives investors more visibility into revenue durability, a higher range becomes easier to defend.[CV023, CV030, CV031, CV032, CV033, CV037]

8.4 Exit readiness and final diligence asks

Hai is not exit-ready on a public-evidence basis yet, but it is close enough to justify focused diligence rather than passive interest. The company has enough commercial proof to clear the “is this real?” hurdle. Official customer and partner materials show production deployments, enterprise use cases, and international channel scale. Market reports show that the category still has structural tailwinds. Those are the reasons to stay engaged. But the missing items are exactly the ones that govern entry price and downside. Public sources still do not provide a clean cap-table bridge, liquidation-preference map, final price range, or the recurring-revenue and cohort disclosures that would let an investor normalize Hai against disclosed public peers. The most useful diligence asks therefore target what would actually change the recommendation. First, get the cap table and the bridge from the last private round to any proposed public pricing. Second, demand segmentation of hardware, software, maintenance, and support economics. Third, test whether concentration is easing or merely being hidden inside growth. Fourth, verify overseas support economics rather than assuming they deserve a premium. Until that package is available, Hai remains a monitor-worthy company with genuine strategic value but incomplete priceability. That makes the current stance decisive but evidence-based: track / research-more, not because the business is weak, but because the valuation inputs that matter most remain only partially public.[CV011, CV014, CV039, CV040, CV041, CV044]

8.5 Exhibits