PaXini Tech

1.1 Identity, business model, and product scope

PaXini Tech (帕西尼感知科技, English legal: PaXini Technology (Shenzhen) Co., Ltd.) was incorporated on 2021-06-29 in Shenzhen, China. The company's registered headquarters is in the Nanshan District at Vanke Cloud City, West Science Street; additional offices operate in Shanghai, Tianjin, and Suqian. The phone and address listings on the official English and Chinese sites confirm all four cities. The company's stated mission is "embodied perception driving general intelligence." Its business model is a full vertical stack: Hall-effect multi-dimensional tactile sensors sit at the base, enabling a dexterous hand product line (the DexH series), which integrates into humanoid robots (the TORA platform). This hardware stack is supported by a proprietary data infrastructure layer—five data collection factories spread across North, East, Central, Southwest, and South China—that produces near-tens-of-billions of multimodal data points annually. PaXini uses this data to train its OmniVTLA full-modality embodied AI model, closing a loop from physical sensor to foundation model with no external data dependency. The core enabling technology is the HAPTA (Hetero-Array Probing Tactile Architecture) and its primary component, the ITPU (Intelligent Tactile Processing Unit). This architecture allows a single sensing unit to capture 15 simultaneous dimensions of tactile information including six-axis force, material texture, temperature, and elastic response at 1000 Hz with 0.01 N full-range resolution. The PX-series sensors span three product series and 12 models, and the company asserts global leadership in 6D Hall-effect sensor shipment volumes. The DexH13 GEN2 dexterous hand—the latest commercially available unit—embeds 1,140 ITPU units across 13 degrees of freedom, 4 fingers, and a 5 kg load capacity with a 15 cm maximum grasp diameter. It carries an 800-megapixel AI hand-eye camera for "tactile + vision" dual-modal perception, which the company claims is an industry first. The TORA-ONE humanoid robot integrates 47 degrees of freedom, 1,956 tactile sensor channels, five monocular cameras, two depth cameras, LiDAR, and laser SLAM. Products are deployed across precision manufacturing, automotive production lines, logistics, healthcare, and consumer electronics. PaXini publicly names BYD and JD.com as strategic deployment partners. The company also participated in CES 2026 in January 2026, where it demonstrated ice-cream preparation by TORA-ONE in the ENTERPRISE AI Zone, drawing media attention and validating commercial readiness. Beyond hardware, PaXini launched—in collaboration with JD Cloud, Tencent Cloud, and Baidu AI Cloud— the world's first large-scale multimodal embodied intelligence data cloud mall, providing standardized data products to the broader industry. The HyperCosmos world model is described by the company as under active R&D, co-funded by the Series B round.[CO001, CO002, CO003, CO004, CO005, CO006]

1.2 Founders, leadership, and governance

The founding team traces its origin to the Sugano Laboratory at Waseda University in Tokyo, which is widely credited as the birthplace of the world's first humanoid robot. Xu Jincheng (许晋诚) holds a doctorate from Waseda and studied directly under Professor Shigeki Sugano, a leading figure in humanoid robotics research. After completing his doctorate, Xu returned to Shenzhen and co-founded PaXini to commercialize force-tactile sensing expertise from the lab. Co-founder Nie Xiangru (聂相如), also known as Mark Nie, holds a Master's degree from the University of Illinois at Urbana-Champaign. His prior career spanned ZTE, Xiaomi, and NetEase, with a focus on large-account operations in the technology sector—bringing enterprise sales and business development capability to complement Xu's technical background. The two founders' combined credentials represent strong founder-market fit: deep academic expertise in tactile robotics combined with commercial go-to- market experience across China's leading technology companies. Xu Jincheng serves concurrently as Chairman of the Board and General Manager (CEO) per the Aiqicha company registry. The board also includes Zhang Hengdi, Chen Mei, Jin Feng, and Wang Li as directors, and Zhang Jinping as Supervisor. These additional board members are named in the Aiqicha corporate filing but their backgrounds are not disclosed in current public sources. Per the QCC (Qichacha) shareholder registry, Xu Jincheng holds approximately 18% of the equity and Nie Xiangru approximately 8%, making them the two largest individual shareholders. BYD (2025-04-22) and JD Technology Information Technology Co., Ltd. (2025-09-12) are also registered shareholders, corroborating media reports of their investment. Key-person dependence is material: Xu Jincheng is the primary technical architect, the public face of the company's academic credentials, and the legal representative. Nie Xiangru is the anchor of the commercial and investor-relations function. Neither appears replaceable in the near term without affecting the company's positioning and investor trust.[CO013, CO014, CO015, CO016, CO017, CO018]

1.3 Funding history, valuation, and investor base

PaXini's capital journey has been rapid even by Chinese embodied AI standards. The first documented external round was a Pre-A closing dated 2022-12-29, described in Baike as raising "tens of millions of CNY" and marking the company's first institutional backing roughly eighteen months after founding. The QCC shareholder registry shows Xin'an Capital (新奥资本) and several other early institutional investors appearing from February 2024 onward, suggesting a Series A ladder began in early 2024. BAIC's investment vehicle (安鹏绿色) is registered from June 2024. BYD entered in April 2025 at the A+ stage, an investment described by Sina Finance as BYD's largest single embodied AI investment in 2025. Yicai reported a 13.4% stake for BYD at the time of the JD-led round in August 2025. On 2025-08-05, JD.com led a Series A financing that brought PaXini's total raised to CNY 1 billion (~USD 139 million) in just four months—a milestone the company announced publicly. That round also included new investors Puxin Capital and XGD (新国都), plus follow-on commitments from existing shareholders TCL Capital and Addor Capital. The Series B closed on 2026-03-06. Multiple independent Chinese and international sources—Equalocean, Gasgoo, and the QQ/猎云网 syndicate—confirm the round exceeded RMB 1 billion and pushed PaXini's post-money valuation above RMB 10 billion (~USD 1.5 billion at prevailing rates). Pandaily's headline described the round at over USD 150 million. The round was led by Huangpu River Capital (黄浦江资本), Kaitai Capital (凯泰资本), and Xin'an Capital (信安资本), with additional participation from a major dual-currency fund, Zhuhai Science and Technology Industry Group, Shanda Capital (善达资本), Haichuan Fund (海川基金), Guangzhou Emerging Fund, the listed company Uni-Trend (优利德), Zhilai Capital (知来资本), LEO LION, Nanling Fund (南岭基金), Xiangcheng Financial Holding (相城金控), and several existing shareholders including Addor/Yida Capital (毅达资本), which significantly increased its stake. The company describes cumulative investment from 14 top-tier global industrial investors including Meta affiliates, BYD, JD.com, TCL, Zhuhai Sci-Tech Industry Group (Huafa and Gree affiliates), SAIC, and BAIC. Tracxn pegs total funding at approximately USD 150 million across four rounds, though its per-round amounts are partially redacted. CB Insights has a profile page but content is gated; no independently verified total has been published outside of company statements and tracker estimates. The pre-Series B JD-led round data from Yicai (a state-owned financial media outlet) is the most reliable anchor for the cumulative CNY 1 billion figure through August 2025.[CO021, CO022, CO023, CO024, CO025, CO026]

1.4 Milestone chronology and adverse context

PaXini's public record provides a clear chronology from founding through its Series B. The company incorporated on 2021-06-29 and released its first ITPU-based tactile sensors and DexH dexterous hand products within its first two years. The Pre-A round (2022-12-29) provided initial institutional capital. The TORA-ONE humanoid robot launched in 2024, integrating the ITPU tactile ecosystem with a full-body robot platform. A Tianjin embodied intelligence data center opened in 2024–2025, and by August 2025, the JD-led Series A brought total capital raised to CNY 1 billion. The US trademark "PAXINI" was filed with Justia on October 14, 2025, signaling a move toward international branding. CES 2026 (January 6–9, 2026) provided the first major international stage for the full product portfolio. The Series B closed on 2026-03-06. Adverse signals require explicit treatment. Aiqicha's company record as of 2026 shows three business operating disputes (经营纠纷), one formal lawsuit filing (立案信息), and one court hearing announcement (开庭公告) against PaXini. These are not immaterial: a court filing and hearing announcement suggest at least one active litigation proceeding. The nature, counterparty, and dollar exposure of these disputes are not disclosed in public filings; direct inquiry with PaXini's legal team is the appropriate diligence path. Two additional adverse signals are traceable from data quality alone. Tracxn reports only 13 employees for PaXini, which directly contradicts the 315 insured employees recorded in Aiqicha's social insurance database. Tracxn's count is almost certainly a LinkedIn-scrape artifact rather than an accurate payroll figure; nonetheless the discrepancy highlights the risk of relying on any single-source headcount proxy. PaXini has not published official headcount figures. Commercialization opacity is a structural gap: despite extensive product coverage and strategic investor involvement from BYD and JD.com, PaXini has disclosed no revenue figure, gross margin, or customer-count data to public sources. Claims of "scale deployment" at automotive production lines and logistics operations are company-sourced and unverified by customer disclosures. Aiqicha classifies PaXini as a 2025 Gazelle Enterprise (瞪羚企业) and Science and Technology SME, suggesting it is recognized as a high-growth early-stage company but not yet at mature scale.[CO033, CO034, CO035, CO036, CO037, CO038]

2.1 Market Boundary and Stack Definition

PaXini's addressable opportunity spans a nested stack of markets rather than a single category. At the base is the robotics-grade tactile and force/torque sensor market—distinct from consumer haptics and medical-only tactile devices, which have different buyer requirements, precision thresholds, and regulatory constraints. Within that base, the high-precision multi-dimensional segment (six-axis force, texture, elastic response captured simultaneously) is narrower still, and is where PaXini differentiates via its Hall-effect technology. Above sensors, PaXini sells dexterous robotic hands (DexH13) that integrate those sensors—competing here against gripper vendors and bare-metal effector suppliers. The next layer is humanoid robot platforms (TORA-ONE, TORA-DOUBLE ONE), where PaXini competes with Unitree, AgiBot, and global players. At the apex is an emerging embodied AI data infrastructure market: the OmniSharing DB, co-hosted with JD Cloud, Tencent Cloud, and Baidu AI Cloud, which commercializes real-world multimodal tactile data. This data layer has no established third-party sizing and should not be included in near-term TAM estimates without explicit caveats. The relevant excluded spend includes consumer haptic actuators (mobile phones, gaming), load cells and uniaxial strain gauges used in industrial weighing, autonomous mobile robots without dexterous arms, and purely software-based robot control stacks. PaXini's value proposition—full-stack hardware plus proprietary dataset plus model—means its SAM crosses all four layers but obtainable penetration in each is at different maturity stages. [CM001, CM002, CM003, CM004]

2.2 Market Sizing Across Multiple Lenses

No single authoritative sizing covers PaXini's precise SAM, so multiple lenses must be triangulated. For the humanoid robot market (the primary demand driver), MarketsandMarkets estimates $2.92 B in 2025 growing to $15.26 B by 2030 at 39.2% CAGR—a paid-access top-down report with broad inclusions. China alone is projected at $0.40 B in 2025 rising to $2.80 B by 2030 at 47.6% CAGR, reflecting the government-backed acceleration described in China's policy actions. The US segment is estimated separately at $857.9 M in 2025 growing to $4.601 B by 2030. In the force sensor market—the industrial robot F/T sensor base on which PaXini's sensors compete—the global market was $2.30 B in 2024 and is expected to reach $3.93 B by 2035 at a modest 5.0% CAGR, reflecting slower growth in conventional industrial applications. The haptic and tactile technology market is broader at $6.61 B in 2025 growing to $8.21 B by 2030 at 4.5% CAGR, but is dominated by consumer haptic devices (vibration motors, gaming controllers) and significantly overstates the robotics-grade sub-segment PaXini targets. For the global tactile sensor market (consumer plus industrial), VMR via Sina Finance reported it exceeded $11.25 B in 2021 and projected it at $26.08 B by 2028, again dominated by non-robot applications. The GGII forecast cited by China Unicom Research is notably more conservative than MarketsandMarkets: 12,400 humanoid units in 2025 at 6.339 B yuan (~$875 M), which is three times smaller than MarketsandMarkets' concurrent global estimate of $2.92 B—likely reflecting a narrower definition limited to advanced bipedal platforms. Goldman Sachs' two cited projections for the 2035 market ($154 B and $38 B) are substantially irreconcilable without access to the primary reports; the $38 B figure from AI Funding Tracker and the $154 B from the WIC report may differ in scope (hardware only vs. full ecosystem), time horizon, or report vintage, and neither should be used as a primary sizing reference without primary access. Investors should treat any single headline TAM with caution: PaXini's near-term SAM in the sensor and hand components of humanoid platforms is plausibly in the $0.3–0.8 B range as of 2025, growing with humanoid unit volumes. [CM009, CM010, CM011, CM012, CM013, CM014]

2.3 Buyer Segmentation and Adoption Path

PaXini's buyer landscape has two structural layers. The first layer is direct commercial customers: manufacturing firms, logistics operators, and healthcare systems that purchase PaXini's sensors, hands, or robots for their own operations. The second—and currently more material—layer is humanoid robot OEM customers: platform builders like Unitree, AgiBot, and Figure that embed PaXini's sensors and hands into their robots and create derived demand. In the OEM channel, the budget owner is the VP of Engineering or CTO, the evaluation criterion is sensing performance (accuracy, repeatability, latency), and the contracting mode is typically a co-development agreement progressing to a volume supply arrangement. In manufacturing verticals, the adoption trigger is economic: rising Chinese labor costs, yield requirements on precision assembly lines (automotive body panels, 3C electronics), and government-mandated automation targets. BYD's investment and customer relationship is the strongest commercial validation available publicly, though the terms and volume are not disclosed. JD.com's logistics deployment validates the warehousing use case, though humanoid robots in that context are still largely at pilot stage industry-wide. Healthcare is the slowest-moving segment: regulatory compliance, liability, and clinical validation requirements extend the sales cycle materially, and few humanoid-grade tactile sensing deployments exist outside R&D settings. Service robotics and retail represent future optionality rather than a 2026 addressable market for PaXini's current price points. The embodied AI data infrastructure segment (OmniSharing DB) adds a software-like recurring revenue dimension but lacks publicly comparable pricing or adoption benchmarks; the data mall with JD Cloud, Tencent Cloud, and Baidu AI Cloud is early stage. [CM021, CM022, CM023, CM024, CM025, CM026]

2.4 Growth Drivers and Policy Tailwinds

Three structural forces are accelerating market growth for PaXini's stack. First, Chinese government policy provides an unusually dense set of demand levers. The 17-ministry "Robot+" Action Plan (January 2023) set explicit targets of 100+ new robot application technologies and 200+ typical scenarios. MIIT's Humanoid Robot Guidance (October 2023) elevated humanoid robotics to a key frontier priority, directing provincial and municipal governments to include embodied AI in industrial upgrading plans. The March 2026 national standard system for humanoid robots—six lifecycle components, developed by 120+ institutions under MIIT—provides compliance clarity that could accelerate procurement in government-adjacent sectors (state-owned manufacturers, public logistics). China's 2025 designation as the first year of mass production, with 140+ domestic manufacturers and 330+ models, confirms the ecosystem is crossing the diffusion threshold. Second, investment flows are self-reinforcing. Global robotics startups raised $13.8 B in 2025, up 77% from $7.8 B in 2024. Humanoid robot funding alone reached a record $1.2 B in 2024 and was projected to double to $2.3 B in 2025. This capital funds OEM customers who are then able to afford higher-quality sensor components. PaXini itself has attracted 14 global industrial investors spanning BYD, JD.com, SAIC, BAIC, and TCL, spanning the automotive, logistics, consumer electronics, and manufacturing supply chains. Third, the AI/LLM integration wave increases the marginal value of rich tactile data. General-purpose humanoid robots need to perform diverse tasks without scene-specific programming; tactile feedback from sensors is critical for safely grasping irregular objects, detecting slip, and adapting grip force in real time. The convergence of LLMs, embodied AI models (VLA architectures), and dexterous manipulation is creating a market for precisely the full-stack infrastructure PaXini provides. [CM029, CM030, CM031, CM032, CM033, CM034]

2.5 Adoption Constraints and Sizing Uncertainty

Several material constraints limit near-term market realization. Most significantly, approximately 68% of the high-end tactile sensor market is dominated by established foreign incumbents—Tekscan (US), Nitta (Japan), and PPS (US)—who have decades of manufacturing know-how, integration histories, and customer relationships in medical and industrial segments. PaXini's claimed 12% domestic market share in high-end tactile sensors (per Sina Finance) leaves substantial foreign incumbent presence, and displacement in premium segments typically requires multi-year qualification cycles. The pilot-to-production gap is the second major constraint. As of mid-2026, most humanoid robot manufacturers are in trial or early deployment phases. TrendForce projects 94% output growth in 2026, but the base is small; Unitree and AgiBot together are expected to account for ~80% of global shipments, indicating a highly concentrated buyer base for PaXini's OEM channel. The unit economics of the DexH13 hand (~$17,200) and TORA-ONE platform ($50,000–$150,000) constrain adoption to capitalized OEM partners and premium industrial applications; broad consumer or SME adoption is several years away at current price points. Integration complexity is a persistent constraint: each OEM deployment requires deep co-development to integrate PaXini's sensors into robot control stacks, and PaXini's headcount limits how many simultaneous integration projects it can support. Finally, the market sizing from analyst firms reflects methodological divergences—GGII's $875 M vs. MarketsandMarkets' $2.92 B for the same 2025 global humanoid market—that should prompt investors to conduct bottom-up sizing using actual unit volume data rather than relying on top-line report figures. [CM036, CM037, CM038, CM039, CM040]

2.6 Exhibits

3.1 Tactile and Force Sensing Incumbent Competitors

Three established companies define the Western tactile/force sensing incumbent segment that PaXini most directly challenges at the component level: Tekscan, Pressure Profile Systems (PPS), and Nitta Corporation. Tekscan (Norwood, MA; privately held) is the category's best-known vendor for pressure mapping and embedded flexible force sensors. Its three core lines— Pressure and Temperature Mapping Systems, Embedded Flexible Force Sensors, and TactileLab custom engineering services—serve R&D engineers performing design validation, QC, and battery-stack testing, as well as clinicians measuring orthopedic and rehabilitation outcomes. Tekscan does not offer a multi-dimensional Hall-effect tactile sensor designed for dexterous-hand OEM integration, and its go-to-market is primarily through R&D equipment channels and direct sales, not robot OEM supply chains. Pricing is not publicly disclosed; system costs vary widely by configuration and are typically obtained via quotation. PPS (Pressure Profile Systems; US headquarters, UK office; privately held) offers capacitive tactile sensors—claiming 25+ years of experience—and its proprietary Chameleon software for pressure visualization. PPS serves 200+ organizations and has explicitly moved into robotics with its TactileGlove, Modular TactileHead, and integrated RoboTact product. PPS argues that capacitive sensing outperforms resistive alternatives in sensitivity, repeatability, and temperature stability. PPS has distribution partners across Asia and offers the RoboTact as an integrated robotics product, making it a more direct competitor than Tekscan. However, PPS is a small company without the full-stack embodied AI integration, humanoid robot body, or proprietary dataset that PaXini offers. Pricing is not publicly disclosed. Nitta Corporation (Osaka; TSE-listed conglomerate) is historically known for force and torque sensors in industrial contexts, but its US division (NCA) focuses on industrial belting, hose/couplings, pneumatic tubing, and RETS tool-changers for robotics—not humanoid tactile sensing. English-language product content from both Nitta's US and Japan English sites covers conveyor and belt products. Nitta's force/torque sensor relevance exists as an industrial adjacency, not as a direct competitor to PaXini's Hall-effect multi-dimensional tactile stack for humanoid dexterous hands. RoboSense (Shenzhen, China; HKEX-listed) is the sharpest single-player threat. Originally a LiDAR and active camera maker for automotive and robotics perception, RoboSense has announced a full-stack pivot that includes a multi-DOF dexterous hand, a VTLA-3D large model for long-horizon task planning, in-house SoC chips (RISC-V data processing, digital SPAD-SoC, 2D addressable VCSELs, MEMS), and a 48,000 m² smart manufacturing facility certified under ISO 26262, IATF 16949, and ASPICE CL2. If RoboSense integrates tactile sensing into its dexterous hand—as its full-stack framing strongly implies—it would offer perception, actuation, and AI in a single supplier relationship, displacing component-level tactile sensor vendors for RoboSense's own robotic platforms and potentially for external OEM customers. [CP001, CP002, CP003, CP004, CP005, CP006]

3.2 Embodied AI, Dexterous Hand, and Humanoid Robot Competitors

The embodied AI and humanoid robot segment is PaXini's largest addressable OEM customer pool and its most dynamic competitive threat vector. The same OEMs that buy PaXini's tactile sensors and dexterous hands could decide to build sensing in-house as they scale to tens of thousands of units annually. Unitree Robotics (Hangzhou; Series C; IPO approved June 2026) leads the Chinese humanoid market by revenue and public profile. The G1 base model is listed at USD 13,500 (excluding tax/shipping); the H1 and H1-2 are larger full-size platforms with optional dexterous hands (Dex3-1 or Dex5-1) and optional tactile sensor arrays. Unitree's 2025 revenue reached 1.699 billion yuan (from 159 million yuan in 2023), with humanoid robots exceeding 51% of revenue and combined gross margins of 60.13%—unusually high for robotics hardware. The STAR Market IPO approval on June 1, 2026 (first "embodied AI" company on China's A-share market) validates Unitree's financial profile and signals plans to expand capacity to 75,000 humanoid and 115,000 quadruped robots annually, requiring massive dexterous-hand and sensor procurement that currently flows through third-party suppliers including PaXini. AGIBOT (Shanghai; 2023; Series B; ~$83.8M raised per Tracxn) produced its 10,000th humanoid robot (Expedition A3) on March 30, 2026, with production accelerating from 1,000 to 5,000 to 10,000 units in approximately three months. Omdia ranked AGIBOT first globally in humanoid robot shipments for 2025. TrendForce projects Unitree and AGIBOT together will capture ~80% of Chinese humanoid shipments in 2026 as output grows 94% year-on-year. AGIBOT has also released AGIBOT World, a large-scale enterprise-quality realistic task dataset and ecosystem for embodied AI, signaling vertical integration ambitions that could extend to sensor sourcing. Deployments span automotive manufacturing, consumer electronics, logistics, and international markets (Europe, Asia, Middle East). UBTech Robotics (Shenzhen; HKEX-listed) targets humanoid service robots for elderly care, AI education, and consumer-grade smart hardware, with its Walker robot as the flagship humanoid platform. As a public company, UBTech is more transparent than most Chinese humanoid peers. UBTech's service focus makes high-fidelity tactile sensing potentially differentiating, but its product pages provide limited technical specification for tactile integration. Figure AI (San Jose, CA; Series C; >$1.9B raised; $39B post-money valuation) is the most funded pure-play Western humanoid company. Its Figure 03 is designed for everyday home environments, powered by the Helix AI system. Figure validated sustained commercial deployment with BMW Group Plant Spartanburg (1,250+ runtime hours, 90,000+ parts, 30,000+ vehicles). The BotQ factory produces one robot every 90 minutes as of early 2026. Figure represents the leading Western benchmark for AI-driven dexterous manipulation in real manufacturing environments. Fourier Intelligence (Shanghai; Series B) develops humanoid robots for rehabilitation and industrial applications through its GR series. However, its official website was inaccessible (SSL certificate error) during this research run, limiting primary-source analysis. LimX Dynamics (Shenzhen; 2022; Series B; $296M raised per Tracxn) is developing embodied intelligent robotics and an agentic operating system. Commercial deployment details and customer announcements are limited in publicly available English-language sources as of June 2026. [CP009, CP010, CP011, CP012, CP013, CP014]

3.3 Feature and Capability Comparison

PaXini differentiates through the depth and integration of its tactile sensing stack rather than the breadth of its robot body portfolio. The critical comparison axes are: (a) multi-dimensional sensing quality, (b) dexterous hand design, (c) humanoid body completeness, (d) proprietary AI model and real-world data platform, and (e) commercial deployment evidence. On tactile sensing quality, PaXini's ITPU (Intelligent Tactile Processing Unit) technology is purpose-built for multi-axis force, texture, and elastic response measurement. The DexH13 integrates 1,140 ITPU units across 4 fingers and 13 DOF, with 5 kg load capacity, 1 mm press positioning accuracy, and an 800MP AI camera for dual-modal visual-tactile perception. The DexH5 entry-level hand offers 9 DOF, 4 kg load capacity, and is listed at approximately USD 2,000 on humanoid.guide. The DexH13 is listed at approximately USD 17,200. By comparison, Tekscan and PPS offer richer per-unit sensing data in controlled R&D settings but do not deliver integrated dexterous hands, AI inference, or humanoid robot bodies. On AI integration, PaXini's OmniVTLA multimodal large model and the accompanying omni-modality data collection infrastructure (near 200 million data points annually from five regional factories) create a closed-loop training pipeline that is structurally difficult to replicate. RoboSense is the only competitor actively building a comparable full-stack AI pipeline (its VTLA-3D model with long-horizon planning for 10+ sequential actions), but RoboSense's starting point is LiDAR/visual perception rather than tactile sensing. On commercial deployment, PaXini's CES 2026 demonstration—TORA-ONE performing autonomous ice-cream making including lever manipulation, ingredient handling, and cup handover—and supply relationships with JD and BYD provide verifiable deployment proof. The PX-6AX-GEN3 sensor covers 3 product series and 12 models with <0.5% full-scale repeatability across 15 sensing dimensions. Vision-only manipulation approaches (LLM-guided grasping without haptic feedback) represent an active substitute, but lack the precise force control for delicate or deformable object handling that tactile sensing enables. Traditional strain-gauge and load-cell sensors (incumbents: Honeywell, Kistler, ATI/Bosch) remain status-quo for industrial robotic force measurement but do not address multi-axis dexterous-hand manipulation. [CP022, CP023, CP024, CP025, CP026, CP027]

3.4 GTM, Pricing, Distribution, and Switching Costs

Go-to-market approaches diverge sharply by company class. Western incumbents (Tekscan, PPS) sell through direct sales and R&D equipment channels; their typical buyer is an engineer, researcher, or clinician—not a robot OEM procurement team. PaXini has repositioned for OEM and platform integrator supply, with direct relationships at JD (logistics robotics) and BYD (automotive), consistent with a B2B sensor-to-OEM model that scales with robot volumes. Pricing transparency is low across the category. Among publicly confirmed data points: the Unitree G1 base model is USD 13,500 (official product page). PaXini's DexH5 is listed at approximately USD 2,000 and the DexH13 at approximately USD 17,200 on third-party product directories. 1X Technologies' NEO home robot is priced at USD 20,000 or USD 499/month (subscription). Tekscan, PPS, Nitta, RoboSense, AGIBOT, UBTech, Fourier, and Figure AI do not publicly disclose product pricing beyond configuration-specific quotes or partner discussions. Switching costs for tactile sensors embedded in a dexterous hand are non-trivial. The ITPU interface, communication protocols (EtherCAT, Modbus), and firmware integration require meaningful engineering effort to migrate to an alternative sensor vendor. PaXini's full-stack SDK support (sensor → hand → AI model → deployment) increases stickiness once integrated. However, modular humanoid robot designs—where arms and hands are intended for replacement—facilitate multi-homing or vendor switching. Tekscan's thin, flexible sensor formats can be retrofitted without redesigning an entire hand structure. Distribution breadth varies significantly. Unitree sells globally via direct e-commerce and distributors, with the G1 publicly available. AGIBOT has announced partnerships in Europe (Germany's Minth Group), Singapore (Singtel Enterprise), Asia, and the Middle East. Figure AI benefits from strategic investors (NVIDIA, Macquarie, Intel Capital) that also act as distribution and channel partners. PaXini's distribution is primarily direct-to-OEM in China, with CES 2026 marking its first major Western exhibition. Western non-federal OEM partnerships beyond JD and BYD are not documented in independent sources reviewed for this chapter. [CP009, CP025, CP026, CP030, CP033, CP034]

3.5 Moat Durability and Competitive Displacement Risk

PaXini's competitive moat rests on three pillars: Hall-effect sensor IP originating from Waseda University's Sugano Lab, a proprietary real-world tactile data pipeline, and deep integration with leading Chinese humanoid OEMs. Each pillar faces distinct displacement risks. The IP moat is real but contestable. Hall-effect force sensing is an established physics principle; alternative multi-axis sensing technologies (capacitive, piezoresistive, optical) are actively developed by Tekscan, PPS, RoboSense, and academic groups. RoboSense's investment in in-house chip design (MEMS, SPAD-SoC, VCSEL) demonstrates that perception-focused competitors can develop custom sensing hardware once they achieve production scale. The data moat is potentially the strongest structural advantage. PaXini's five-factory omni-modality data acquisition network—with claimed capacity for nearly 200 million data points annually and plans to offer cloud access globally—creates a training advantage for OmniVTLA that is capital- and time-intensive to replicate. However, AGIBOT World (a competing enterprise-quality dataset for embodied AI) and academic open-source datasets represent realistic alternatives; the quality and diversity differential has not been independently benchmarked. The OEM customer moat depends on the ongoing willingness of Unitree, AGIBOT, and other humanoid OEMs to remain component buyers rather than vertically integrate. TrendForce reports Unitree plans to expand capacity to 75,000 humanoid robots annually—a volume at which in-house sensing development becomes economically attractive. A post-IPO acquisition of a tactile sensor company by Unitree or AGIBOT would be a material adverse event for PaXini. The broader Chinese humanoid sector carries legislative and geopolitical risk for Western distribution. The American Security Robotics Act (March 2026, Senators Cotton and Schumer; House companion bill from Rep. Stefanik) would bar U.S. federal agencies from purchasing humanoid robots tied to foreign adversaries, naming Unitree and AGIBOT specifically. China's National Intelligence Law (Article 7, 2017) requires all Chinese organizations to cooperate with state intelligence agencies regardless of data location or subsidiary structure—a structural compliance concern for Western buyers of Chinese robotic components including PaXini's sensors and hands. A Morgan Stanley buyer survey found only 23% of prospective industrial buyers report satisfaction with currently available humanoid products, suggesting that supply-side enthusiasm has outpaced real demand verification. The UniPwn Bluetooth vulnerability disclosed in September 2025 for Unitree robots (enabling full device takeover and automatic spread to nearby robots) demonstrates the systemic security risks embedded in the Chinese humanoid platform ecosystem on which PaXini's customer base is concentrated. [CP015, CP016, CP027, CP028, CP030, CP031]

3.6 Exhibits

4.1 Revenue Model and Monetization Strategy

PaXini operates a three-tier hardware-first revenue model anchored in B2B enterprise sales: tactile sensor components, dexterous hand assemblies, and full humanoid robot platforms. The company targets industrial OEMs, automotive manufacturers, logistics operators, healthcare institutions, and electronics assemblers through direct strategic partnerships—principally BYD, JD.com, SAIC, BAIC, and Geely—rather than through a traditional channel or reseller network. Revenue is booked when hardware is delivered to the enterprise customer; realized pricing, volume discounts, and contract structures are not publicly disclosed. A fourth revenue stream is in active development: PaXini plans to open its omni-modality embodied AI data globally via a cloud data store, monetizing its claimed output of ~200 million data entries per year. This data monetization pathway was not commercially active as of June 2026, and no launch timeline or pricing model has been disclosed. Professional integration and support services are mentioned in product literature as part of the enterprise offering, but no standalone service revenue breakdown is available. The company's investor roster—spanning internet giants (JD.com, Meta affiliate LEO LION), automotive OEMs (BYD, SAIC, BAIC, Geely), electronics (TCL, Uni-Trend), and state-owned industrial funds—signals an intent to monetize through deep strategic embedding rather than transactional volume, which carries customer-concentration risk that cannot be quantified from public sources. Revenue recognition policies are not disclosed; hardware product lines likely follow standard delivery-based recognition, but enterprise robot deployments may use milestone-based models given the integration complexity.[CI001, CI002, CI003, CI004, CI043, CI044]

4.2 Pricing Architecture and Competitive Context

Product pricing data for PaXini is available only through third-party marketplace aggregators and analyst commentary, not from official PaXini pricing pages. The DexH5 entry dexterous hand is listed at approximately $2,000 per unit on Humanoid.Guide; the DexH13 Gen 2, a more capable four-finger hand with 1,140 ITPU tactile processing units, is listed at approximately $17,200. The Tora-One humanoid robot system carries an estimated range of USD $50,000–$150,000 based on RBTX marketplace quotes and institutional pricing signals on OriginOfBots—list pricing only, with realized enterprise discounts unknown. At the sensor component level, CNEVPost citing Goldman Sachs research attributes a sensor unit price as low as RMB 199 (~$28) to PaXini's multi-dimensional tactile sensors, representing a dramatic cost disruption of what had been a laboratory-only technology. Goldman Sachs also reportedly attributes >80% of the global robotic tactile sensor market to PaXini, making it a core supplier—a claim that originated in an analyst research report not independently accessible to this analysis and should be treated as third-party attributed rather than verified. Competitor pricing context: Unitree's G1 humanoid robot and H1 platform are positioned as lower-cost consumer-oriented systems, while PaXini's Tora-One targets industrial precision tasks at a higher ASP. The wide spread between DexH5 ($2,000) and DexH13 ($17,200) implies tiered pricing for different robot platform integrations. All third-party pricing sources should be treated as marketplace estimates; realized enterprise pricing, volume discounts, and multi-year contracts are undisclosed diligence requirements.[CI005, CI006, CI007, CI008, CI009, CI010]

4.3 Cost Structure and Capital Intensity

PaXini's cost structure is hardware-intensive and compounded by a second capital sink: large-scale AI data collection infrastructure. Major cost categories include: (1) proprietary sensor chip and ITPU module development, (2) sensor packaging and manufacturing scale-up, (3) five regional omni-modality data collection factory clusters across China (North, East, Central, Southwest, and South China), (4) GPU compute for OmniVTLA and HyperCosmos model training, and (5) a headcount base of 315 insured employees (AiQiCha social insurance records) against an enterprise-grade talent profile from Waseda University robotics lineage. The Tracxn-reported 13-employee figure is likely a stale index artifact and conflicts with the AiQiCha registry evidence at an order of magnitude; the 315 figure is better supported. No gross margin, COGS breakdown, R&D expenditure, or operating expense data is publicly disclosed. The most useful industry benchmark is Unitree Robotics's FY2025 filing from its STAR Market IPO prospectus (cited by TrendForce), which reports a 60% combined gross margin across humanoid and quadruped robots—a hardware robotics company achieving this margin suggests high-volume manufacturing scale is achievable, but PaXini is at an earlier commercialization stage. PaXini's additional data factory capex has no Unitree parallel, creating a structurally heavier cost base than pure robot OEMs. Without COGS or margin data, the capex intensity of the five data cluster facilities cannot be estimated. Monthly cash consumption is expected to be material given the multi-site physical infrastructure, headcount, and parallel model development programs, but the magnitude is an irreducible diligence gap.[CI019, CI020, CI021, CI022, CI023, CI024]

4.4 Capital Adequacy and Funding Dependency

PaXini closed its Series B in March 2026, raising more than RMB 1 billion (approximately $144.93–$150 million USD, with minor source variation in aggregation methodology). The round was led by Huangpu River Capital, Kaitai Capital, and state-backed Xin'an Capital, with 13+ co-investors including Zhuhai Science and Technology Industry Group, Shanda Capital, Haichuan Fund, Guangzhou Emerging Industry Fund, Uni-Trend, Zhilai Capital, LEO LION, Nanling Fund, and Xiangcheng Financial Holding. Existing shareholders—including Addor Capital, Yida Capital, and TCL Capital—significantly increased their stakes. The total cumulative raise is $144.93 million per CB Insights (9 rounds) or $150 million per Tracxn (4 rounds aggregated differently); both figures reflect capital raised through March 2026 and the discrepancy stems from sub-round aggregation methodology. Cash on hand, monthly burn rate, and remaining runway are not disclosed; no estimates can be made without data room access. The company has earmarked the Series B proceeds for: (i) hardware upgrade of proprietary perception systems, (ii) expansion of its super data collection factories toward ~200 million data entries per year, (iii) R&D on OmniVTLA and HyperCosmos AI models, and (iv) commercial pilot expansion and scale deployment. PaXini is additionally exploring a Hong Kong IPO (Bloomberg, via CNEVPost, June 2026), working with banks on a potential listing application in the coming months; successful execution would provide material capital for further expansion. The company's funding trajectory—nine rounds in under three years, with round size escalating from seed to >$140M Series B—reflects structural dependence on external venture and strategic capital, with the IPO as the next planned liquidity and growth-capital event. Deal terms for the Series B (pre-money valuation, liquidation preferences, conversion prices, anti-dilution provisions) are all marked non-public on CB Insights.[CI011, CI012, CI013, CI014, CI015, CI016]

4.5 Financial Gaps, Adverse Evidence, and Diligence Verdict

PaXini's financial opacity is the defining diligence risk at this stage. The company has not disclosed revenue, ARR, GMV, gross margin, burn rate, runway, CAC, LTV, payback period, or any per-customer revenue metric. CB Insights records zero revenue for all nine funding events, confirming the absence of any verified external financial reporting. A Bloomberg-sourced IPO inquiry (reported by CNEVPost, June 2026) did not include financial disclosures. Adverse evidence from the Chinese corporate registry (AiQiCha) reveals three recorded business disputes, one filed legal case, and one court hearing announcement—the nature, counterparty, and financial materiality of these proceedings are not disclosed and require independent legal due diligence. The >80% global tactile sensor market share attributed to PaXini by Goldman Sachs originates from a single analyst report cited through a single media outlet (CNEVPost); it has not been corroborated by independent market data and should not be used as the primary basis for market-sizing assumptions. Revenue quality risk is high: hardware sales to a small cluster of strategic investors who also happen to be customers (BYD, JD.com, Geely) create a related-party revenue concern that cannot be evaluated without full financial statements. All third-party pricing data—DexH5 $2,000, DexH13 $17,200, Tora-One $50K–$150K—is derived from marketplace aggregators, not from official PaXini pricing pages or contracts. The financial diligence verdict is: the company cannot be underwritten on public data alone. Material blockers include: (1) no disclosed revenue or margins, (2) no burn rate or runway, (3) related-party revenue concentration risk, (4) unquantified legal proceedings, and (5) IPO-level capital dependency for the next growth phase. The Geely deployment (99.8% task success, 8-hour continuous operation) is promising evidence of technical readiness, but is self-reported and requires independent audit.[CI025, CI027, CI028, CI029, CI030, CI033]

4.6 Exhibits

5.1 Product Line, SKU Map, and Customer Workflow

PaXini's commercial portfolio is organized across three tiers that form a closed loop: (1) tactile sensors and sensing units sold to robot integrators and OEMs; (2) dexterous hand and humanoid robot platforms that integrate those sensors into end-effectors and full-body systems; and (3) an embodied AI data infrastructure that captures real-world interaction data and feeds proprietary AI models. The company describes this as the "Tactile Sensors – Robot Platform – Embodied Data – Scenario Application" closed loop, positioning PaXini as a full-stack infrastructure provider rather than a point-solution sensor vendor. Customers engage PaXini at different entry points: pure sensor buyers (robot OEMs, university labs) purchase ITPU units or the PX-6AX-GEN3 series for integration into third-party arms and grippers; dexterous-hand buyers (humanoid robot makers, logistics integrators) acquire DexH5 or DexH13 as ready-to-integrate end-effectors; and humanoid-platform buyers (large industrial and healthcare accounts such as JD and BYD) deploy TORA-ONE for specific tasks such as automotive assembly and goods handling. A separate AI developer segment accesses the OmniSharing DB cloud data mall for training data. All tiers are intended to reinforce each other: sensor sales fund hardware IP, robot deployments generate proprietary interaction data, and that data trains the OmniVTLA model. [CE001, CE002, CE003, CE038, CE039, CE051]

5.2 ITPU/HAPTA Tactile Architecture and Sensor Specifications

PaXini's core sensing technology is built on Hall-effect multi-dimensional sensing physics, in which an array of Hall-effect magnetic sensing elements converts applied force into measurable voltage changes across multiple spatial directions simultaneously. This physical layer is organized within the HAPTA (Hetero-Array Probing Tactile Architecture), a heterogeneous multi-array software and hardware architecture that integrates ITPU sensing units with onboard tactile algorithms to enable millisecond-level signal processing. The ITPU (Intelligent Tactile Processing Unit) is the discrete functional sensing block embedded in all PaXini products. The production sensor, the PX-6AX-GEN3 series, spans three sub-series and 12 models and measures 15 sensing dimensions simultaneously including six-axis force (Fx, Fy, Fz, Tx, Ty, Tz), material texture, and elastic response. Per company-published specifications corroborated by a third-party product listing, the ITPU array supports up to 121 force signals per array, a normal force range of 0–25 N and tangential range of ±10 N, minimum recognition force of 0.01 N, spatial resolution of 0.1 mm, sampling rate of 1 MHz (maximum output frequency 1,000 Hz), and repeatability of <0.5%FS. Environmental ratings include IP65, UL94 V-0 flame retardant certification, 200% safe overload, 300% shock overload, chip-level stray magnetic field immunity, and service life exceeding 10 million cycles. The entry-level sensor lists from US$49, which PaXini positions as a cost reduction for the embodied AI developer market. The PX6D/PXTS is described by PaXini as the world's first commercial Hall-effect 6D force/torque sensor for embodied AI, targeting wrist and joint force control on robot arms. It combines extreme lightweight design with on-sensor intelligence for disruptive cost-performance in whole-body force perception. This claim has not been independently verified by a third-party reviewer as of the run date. [CE004, CE005, CE006, CE007, CE008, CE009]

5.3 Dexterous Hand Platform — DexH5 and DexH13

PaXini produces two commercially available dexterous hand models that embed ITPU sensing directly into the finger structure. The DexH5 (GEN1) is a four-finger, nine-degrees-of-freedom hand with integrated multi-dimensional tactile sensors targeting service robots, research platforms, and light industrial applications. It provides 4 kg payload and uses lightweight high-strength composite materials. A modular wrist interface enables attachment to various robotic arm platforms. The DexH5 represents PaXini's earlier generation, and public specification documentation is limited relative to the newer GEN2. The DexH13 (GEN2) is the current flagship dexterous hand. Its key architecture is the combination of 1,140 ITPU GEN2 tactile processing units (GEN2 generation) generating 3,420 signal channels across four fingers and 13 degrees of freedom, coupled with an 800-megapixel AI camera using zero-shot pose estimation for object recognition without prior training on that object. The company-claimed specs include 15 N fingertip force, 5 kg payload capacity, 15 cm maximum grasp diameter, and under 1.0 s minimum open-close cycle time. For industrial reliability, the DexH13 is rated at more than 1 million press cycles. PaXini describes DexH13 as the industry's first dual-modal dexterous hand combining multidimensional tactile sensing and AI vision. Integration is enabled through standard EtherCAT and Modbus industrial communication protocols. At CES 2026, DexH13 demonstrated stable grasping of irregular objects including test tubes and cubes, and performed delicate knob-turning tasks; independent assessors noted the demonstration but did not publish controlled benchmark results. [CE014, CE015, CE016, CE017, CE018, CE019]

5.4 Humanoid Robot Platform — TORA-ONE and TORA-DOUBLE ONE

PaXini's humanoid robot line currently comprises two models: TORA-ONE (wheeled) and TORA-DOUBLE ONE (bipedal), both integrating the ITPU-based tactile sensing ecosystem. TORA-ONE is a wheeled humanoid platform with adjustable height from 1.46 to 1.86 meters and 47 total degrees of freedom (21 in the body, 26 in its four-fingered bionic hands). The tactile specification is 1,956 sensors across 7,824 channels enabling multidimensional touch perception across the body surface. Navigation and perception are provided by five monocular cameras, two depth cameras, LiDAR, and laser SLAM for 360-degree navigation and 3D mapping. Company-claimed performance metrics include ±0.05 mm arm positioning accuracy, 8 kg load capacity per arm, and 8-hour battery life with real-time visual-tactile fusion. Pricing is estimated at USD 50,000–150,000 based on third-party market listing data and institutional quotes, not an official list price confirmed by PaXini. TORA-ONE was launched in 2024 and demonstrated autonomous task execution including a complete ice-cream making process at CES 2026. TORA-DOUBLE ONE is a bipedal variant shown at CES 2026 demonstrating advanced obstacle-crossing motion control and visitor handshake interactions. Detailed technical specifications for TORA-DOUBLE ONE are not publicly disclosed as of the run date; the platform appears to be at a demonstration or pre-commercial stage. Both models are positioned for industrial manufacturing, automotive assembly, logistics, healthcare/rehabilitation, and commercial service scenarios, based on PaXini marketing materials and partner announcements with JD, BYD, and Geely. [CE022, CE023, CE024, CE025, CE026, CE027]

5.5 Embodied AI Data Infrastructure and OmniVTLA Model

PaXini has built a "human-centric" omni-modality embodied AI data infrastructure it claims addresses the scarcity of physical-contact modality training data in the robotics industry. The physical infrastructure consists of five wholly self-built data collection factories spanning North, East, Central, Southwest, and South China, with claimed annual production capacity of nearly 200 million omni-modality data entries. Data collection is performed using the PXCap V, a wearable device with full five-finger tactile coverage incorporating 30 six-dimensional tactile modules at a density of 3,015 sensing points, capable of simultaneously capturing 82 degrees of freedom of hand motion and 15 dimensions of tactile information (forces, texture, elasticity). The data format is compatible with mainstream frameworks and supports conversion to LeRobot standard format. In March 2026, PaXini launched the OmniSharing DB cloud data mall in collaboration with JD Cloud, Tencent Cloud, and Baidu AI Cloud, described as the world's first large-scale multimodal embodied intelligence data cloud. The platform enables one-stop online data purchase and cloud-based preprocessing, model training, and deployment across the three partner cloud platforms. The proprietary OmniVTLA multimodal large model integrates Visual, Tactile, Language, and Action data streams for embodied AI training. External validation is limited to a joint paper titled "RoboPaint: From Human Demonstration to Any Robot and Any View," co-authored with Shanghai Jiao Tong University and Zhejiang University, which reported 84% success rate in mapping human hand states to robot dexterous hands across 10 tasks. VLA policies (Pi0.5) trained solely on PaXini- generated data achieved an average 80% success rate in three representative tasks within the Isaac Sim simulation environment. These results are simulation-only and do not constitute independent real-world validation of the OmniVTLA model or the data pipeline. PaXini also announced HyperCosmos, a multimodal world model under development, and plans to open the OmniSharing DB globally as a data infrastructure resource. [CE029, CE030, CE031, CE032, CE033, CE034]

5.6 Integration, Reliability, Support, and Roadmap

Integration with third-party robot platforms is facilitated at the hardware layer through standard EtherCAT and Modbus industrial communication protocols on the DexH13 hand. The OmniSharing DB data format supports conversion to LeRobot and is compatible with mainstream deep learning frameworks, enabling AI developers to consume PaXini training data without custom preprocessing pipelines. The company provides both domestic (mkt@paxini.com) and overseas sales channels (sales_global@paxini.com) and maintains offices in Shenzhen (HQ), Shanghai, Tianjin, and Suqian. SPiRAL Labs is listed as a named research partnership on the official website. Published reliability specs include IP65 water/dust resistance, UL94 V-0 flame certification, and chip-level magnetic interference immunity on ITPU sensors. Operating temperature range is 0–50°C, with overload tolerance to −20–120°C. The DexH13 is rated for over 1 million press cycles and the ITPU sensor for over 10 million. These figures are company-claimed; no third-party certification body or independent test laboratory is identified in public materials. The publicly visible roadmap includes: (1) continued DexH13 GEN2 scale production; (2) OmniVTLA model advancement; (3) HyperCosmos multimodal world model development; (4) global expansion of the OmniSharing DB data cloud; and (5) expansion of data collection factory capacity. Specific timelines, milestone gates, or committed customer delivery dates are not publicly disclosed. [CE040, CE041, CE042, CE043, CE044]

5.7 Differentiation, IP, Supply Chain, and Trust/Safety/Quality

PaXini's claimed differentiation rests on four interlocking advantages: (1) proprietary Hall-effect HAPTA/ITPU sensing technology that the company states breaks through "import reliance" in high-end tactile sensors for Chinese robotics; (2) the founding team's academic pedigree from the Sugano Laboratory at Waseda University, which built the world's first humanoid robot; (3) a closed-loop data flywheel—the only company globally, per its own and EqualOcean's characterization, that integrates large-scale tactile modality data into full-modal embodied datasets for training; and (4) the full-stack product matrix from sensor to robot to AI model. Intellectual property: PaXini's Chinese business registry records show 135 patents, 43 software copyrights, and 192 registered trademarks. A USPTO trademark filing for PAXINI (Serial No. 99442343) was submitted in October 2025, covering categories including robotic lawn mowers and beverage processing machines. The patent portfolio is not individually enumerated in public sources and the mix of utility patents, utility-model patents, and design patents is not specified. Supply chain: PaXini manufactures products from its Shenzhen headquarters with additional offices in Shanghai, Tianjin, and Suqian. Specific component suppliers for Hall-effect magnetic elements, hollow-cup motors (used in DexH13 joints), custom PCB/ASIC chips, and structural materials are not publicly disclosed. This represents a material diligence gap for assessing single-source risk and supply continuity. Trust and safety: Published safety specs include IP65 and UL94 V-0 (sensor), chip-level anti- magnetic immunity, and 200%/300% safe/shock overload ratings. The Chinese business registry (AiQiCha) records 3 business disputes, 1 filing record, and 1 court announcement for PaXini Technology (Shenzhen) Co., Ltd.; the nature and materiality of these proceedings are not publicly detailed. [CE045, CE046, CE047, CE048, CE049, CE050]

5.8 Exhibits

6.1 Customer Base Segmentation

PaXini addresses six identifiable buyer segments across its product stack. First, humanoid and embodied-AI robot makers integrate PaXini's ITPU sensors, DexH13/DexH5 dexterous hands, or full robot platforms into their own systems; Figure AI is mentioned in Chinese investment media as an international preferred partner, though no contract has been confirmed. Second, automotive OEMs—primarily BYD, Geely, SAIC, and BAIC, all also equity investors—deploy TORA-ONE humanoid robots on production lines for precision tasks such as wiring-harness assembly. Third, industrial manufacturing integrators in 3C electronics and precision assembly use PaXini's force-torque sensors for automation tasks where sensor entry prices start as low as CNY 199 (~USD 27). Fourth, cloud and data platforms—JD Cloud, Tencent Cloud, and Baidu AI Cloud—act as infrastructure partners and data-monetization channels for the OmniSharing DB data mall. Fifth, academic and research institutions such as SJTU and Zhejiang University purchase sensors and data for embodied-AI model development. Sixth, healthcare, elderly-care, and commercial service operators are targeted by TORA-ONE's mobility and dexterity capabilities, though no named healthcare deployment has been announced. PaXini's investor roster—BYD, JD.com, SAIC, BAIC, TCL, LEO LION (Meta affiliate)—also serves as a proxy for the target customer base, since most investors have disclosed deployment plans or strategic alignment. Geographic coverage remains overwhelmingly China-domestic, with offices in Shenzhen, Shanghai, Tianjin, and Suqian, though the international storefront and AIFITLAB distributor channel provide a global entry point for developer and researcher buyers at the sensor level.[CU001, CU013, CU015, CU016, CU029, CU033]

6.2 Named Customer Deployments and Evidence Quality

The strongest available deployment evidence is Geely Auto's Qiantang manufacturing base, where PaXini's TORA-ONE was deployed in May 2026 for automotive wiring-harness assembly and adhesive-pasting tasks. The company reported eight hours of continuous zero-anomaly operation and a 99.8% task success rate, though these figures originate from a company-issued press release and have not been independently audited. BYD made a strategic investment in April 2025—reportedly its largest single robotics investment—and has co-developed production-line customization requirements with PaXini, but no BYD deployment KPIs have been publicly disclosed. JD.com led the Series A round in August 2025 and has committed verbally to logistics and warehousing deployment via JD AI Research statements, but no JD facility deployment has been announced. SPiRAL Labs is listed as a named customer on PaXini's official website, indicating at least one US-market customer engagement, though scope and production status are unavailable. The joint OmniSharing DB data mall with JD Cloud, Tencent Cloud, and Baidu AI Cloud represents a live partner-customer co-deployment where the three cloud companies serve as both distribution infrastructure and early adopters of PaXini's embodied data layer. Academic partners SJTU and Zhejiang University co-authored the RoboPaint validation paper, confirming research-grade use of PaXini's tactile dataset and dexterous-hand products. SAIC Motor and BAIC Group are early investors with automotive-sector alignment, but neither has disclosed a production deployment.[CU002, CU003, CU004, CU005, CU006, CU007]

6.3 Adoption Trajectory and Market Evidence

PaXini claims industry-leading sensor shipment volumes without disclosing unit counts or year-over-year growth, making adoption trajectory assessment indirect. The most-cited proxy is a Goldman Sachs statement that PaXini has captured over 80% of the global robotic tactile sensor market; however, this figure originates from Goldman Sachs analyst research likely informed by company-provided data, and the scope—whether it covers tactile sensors deployed specifically in humanoid robots or the broader robotics market—is not defined in available sources. A Sina Finance competitive analysis published independently in April 2025 cited PaXini's domestic high-end tactile sensor share at approximately 12%, in a segment where 68% of share is held by foreign incumbents, presenting a materially different picture of competitive position. CES 2026 served as PaXini's largest international proof-of-concept event, where the full product matrix—sensors, DexH13, TORA-ONE, and OmniSharing DB—was demonstrated live. The company's five data factories are designed for approximately 200 million omni-modality data entries annually, and joint research with SJTU and Zhejiang University validated an 84% success rate in human-to-robot hand mapping and 80% VLA policy success in Isaac Sim environments. PaXini is one of eight Chinese embodied-intelligence companies valued above RMB 10 billion, a peer benchmark underscoring the sector's momentum. The IFR 2025 report shows China installed 295,000 industrial robots in 2024—54% of global installations—providing a strong macro tailwind for domestic customer acquisition even absent PaXini-specific adoption data.[CU011, CU012, CU024, CU025, CU027, CU034]

6.4 Retention, Durability, and Expansion

PaXini has not publicly disclosed net revenue retention, gross retention, churn rates, contract lengths, repeat purchase rates, or cohort-level customer satisfaction data. The business model spans one-time hardware sales—sensors at CNY 199–thousands and TORA-ONE at USD 50,000–150,000 per unit—and potentially recurring data subscription revenue via OmniSharing DB, but the data mall's subscriber count and pricing terms are not disclosed, making durability assessment speculative. The structural retention driver is ecosystem lock-in: customers integrating PaXini's ITPU tactile hardware are dependent on compatible dexterous hand models and, ultimately, the OmniVTLA training pipeline, creating an incentive to deepen the relationship over time. Investor-customers such as BYD and JD create long-term strategic alignment but make independent retention measurement difficult because these relationships blend equity interest with commercial obligation. PaXini maintains four China offices (Shenzhen, Shanghai, Tianjin, Suqian) indicating domestic geographic expansion; international reach is nascent, with AIFITLAB as a confirmed third-party global distributor. Korean technology conglomerates are mentioned as international partners in Chinese investment media, but no named Korean entity or revenue has been confirmed. The company's 315 insured employees per the QCC Chinese business registry suggests a lean workforce relative to its RMB 10 billion valuation, which may constrain its ability to service a rapidly expanding enterprise customer base without material hiring investment.[CU014, CU018, CU022, CU026, CU028, CU039]

6.5 Concentration and Partner Dependence

PaXini's customer concentration risk is structurally elevated because its principal named production-intent customers—BYD, JD.com, Geely, SAIC, BAIC—are also equity shareholders. The 14 disclosed top-tier industrial investors span automotive (BYD, SAIC, BAIC, Gree-adjacent Zhuhai Sci-Tech), internet and logistics (JD.com), consumer electronics (TCL, XGD), and overseas (Meta affiliate LEO LION), creating a customer pipeline that is predefined by the cap table rather than purely won through market competition. This structure provides strong near-term commercial validation but creates concentration risk if any anchor partner reduces strategic commitment, and potential procurement-integrity questions if arm's-length pricing is not demonstrable. Goldman Sachs and Morgan Stanley have publicly named PaXini as upstream infrastructure for embodied intelligence, adding reputational weight but also underscoring the circularity risk: analyst market-share claims may derive from company-provided figures without independent verification. The 80% global share assertion conflicts with the Sina Finance 12% domestic high-end share figure; the discrepancy is unresolved and represents a material diligence gap for any investor attempting to size the addressable customer base independently. PaXini's dependence on China's automotive and robotics manufacturing ecosystem also means that regulatory changes, export controls, or sectoral downturns could disproportionately affect the top customer concentration without an international backstop in place.[CU017, CU019, CU023, CU030, CU031, CU037]

6.6 Exhibits

7.1 Regulatory & Legal Risks

PaXini's regulatory risk profile is shaped by three reinforcing forces: the structural obligations of Chinese corporate domicile, an evolving domestic safety and standards regime, and foreign legislative pressure that could progressively close international markets. China's National Intelligence Law (Article 7, 2017) requires all Chinese companies and citizens to cooperate with state intelligence agencies on demand, regardless of server location, corporate structure, or stated privacy policy. For a company collecting rich tactile and multimodal embodied AI data across industrial deployments—and building large-model infrastructure from that data—this obligation creates a structural barrier with customers and governments that restrict foreign-intelligence-accessible systems. This is not an operational risk to be mitigated; it is the applicable law of PaXini's home jurisdiction. US federal agencies, defense contractors, and their supply chains face the most acute exposure, and legislative momentum is building toward formal restrictions. In March 2026, Senators Cotton and Schumer introduced the American Security Robotics Act (with a House companion from Representative Stefanik), which would bar federal agencies from Chinese humanoid robots; peer companies Unitree and AgiBot are cited by sponsors, and the sector context directly applies to PaXini. A companion Humanoid ROBOT Act (November 2025) extends proposed restrictions to federal contractors. Neither bill has passed, but the legislative record signals durable pressure that will constrain PaXini's US government and defense-contractor market even before formal enactment. China's domestic regulatory timeline is simultaneously tightening. MIIT's October 2023 Humanoid Robot Innovation Development Guidance set a 2027 target for a safe, reliable, and controllable industrial chain. In March 2026, the Ministry released China's first national standard system for humanoid robots—developed by 120+ institutions—encompassing safety, ethics, lifecycle compliance, and embodied AI model data governance. Early movers that have embedded hardware in customer facilities before standards solidified face redesign risk if non-conforming components require replacement under future certification regimes. The MIIT guidance explicitly prioritises 3C electronics and automotive sectors as lead applications, precisely the domains in which PaXini's Geely deployment sits, creating regulatory scrutiny concentration. PaXini's own legal record contains disclosed but undescribed adverse items. Aiqicha and QCC data record three business disputes (经营纠纷), one formal case filing (立案信息), and one court session announcement (开庭公告). The nature—contractual, employment, IP, trade-secret, or customer dispute—is not publicly available. PaXini's US trademark filing (October 2025 per Justia) and its 135-patent, 43-software-copyright, 192-trademark domestic portfolio suggest an active but recently assembled IP defence. IP ownership provenance from the Waseda University founding context has not been independently documented. [CR001, CR002, CR003, CR004, CR005, CR006]

7.2 Operational, Technical & Safety Risks

PaXini's commercial narrative depends on hardware that performs reliably at industrial scale, but the current evidence base is thin. Company-specified durability metrics—more than 100,000 operation cycles for the DexH13 dexterous hand and over 10 million service-life cycles, IP65 protection, and 200% safe overload tolerance for the PX-6AX-GEN3 sensor array—have not been independently validated in sustained production environments. The sole confirmed production proof point, Tora-One's deployment at Geely's Qiantang manufacturing base, reported 99.8% task success over eight continuous hours per a company statement in May 2026. A single-facility, single-shift measurement from one deployment is insufficient to constitute a reliability dataset for capital equipment intended for industrial scale. Sector-wide constraints compound PaXini's hardware risk. A Morgan Stanley survey of prospective industrial buyers found only 23% satisfaction with currently available humanoid robot products—a demand-side reality check that aggressive supply-side forecasts do not adequately reflect. Battery life of two to three hours per charge is an industry-level constraint that limits shift coverage, adds hot-swap logistical complexity, and restricts deployment contexts. Most humanoid robot shipments globally through 2025 remained concentrated in controlled pilots and demonstrations rather than continuous multi-shift production; PaXini's Geely deployment is among the clearest documented exceptions, but rests on a company statement, not independent audit. Product safety liability exposure is material in an industrial context. If a Tora-One or DexH13 unit caused injury or production loss at a customer facility, PaXini would face product liability claims without a documented safety audit history or ISO certification track record. The March 2026 national standard system explicitly includes safety-and-ethics requirements spanning the entire industrial lifecycle; non-conformance with evolving standards could trigger enforcement action and customer contract termination. The five-region data collection factory network also introduces operational complexity: sensor calibration, data-quality assurance, and security across distributed facilities demand management bandwidth that is currently unquantified. AI model failure or misclassification in safety-critical operations—PaXini's OmniVTLA 2.0 is a first-generation commercial model— represents an emerging liability with no disclosed red-teaming or fail-safe protocol. [CR010, CR011, CR012, CR013, CR014, CR015]

7.3 Partner, Dependency & Supply Chain Risks

PaXini's strategic architecture creates dangerously concentrated single-counterparty dependencies across its investor, customer, and commercial-validation dimensions simultaneously. BYD made a strategic investment in April 2025—described as BYD's largest single robotics investment and its first major embodied intelligence commitment—and is simultaneously a target customer for industrial automation. JD.com invested through its JoyGen vehicle in Series A III (August 2025) and is a probable commercial reference account. Both investors hold dual roles as capital providers and potential customers, meaning any deterioration in those commercial relationships—from competitive product launches, technology pivots, or geopolitical factors—would impair both PaXini's capital base and its customer proof simultaneously. Commercial concentration is acute. The Geely Qiantang deployment, announced May 2026, is the only confirmed full-production-grade reference. A single-site, single-customer proof point underpins the entire industrial deployment narrative at the time of the planned IPO filing. Failure to extend to additional independent production deployments before the HK IPO would leave incoming public-market investors underwriting one manufacturing facility as the commercial evidence base at a greater than RMB 10B valuation. Supply-chain opacity is a structural gap. PaXini's Hall-effect sensing chip design and ITPU (Intelligent Tactile Processing Unit) technology rely on component manufacturing that is not publicly disclosed. The company describes its sensor cost structure as disruptive— reducing multi-dimensional tactile sensor unit prices to as low as RMB 199—but the underlying component sourcing, margin, and supplier concentration are unknown. Competitor RoboSense has launched a competing full-stack embodied intelligence platform including dexterous hands and IATF 16949- and ISO 26262-certified manufacturing operations, increasing competitive pressure on PaXini's supply-chain advantage. TrendForce projects Unitree and AgiBot will together capture nearly 80% of Chinese humanoid robot shipments in 2026, compressing the market available to component-layer and full-stack humanoid suppliers. The cap table's 50-plus shareholder entities across 38-plus investors also present governance complexity, with liquidation preferences and anti-dilution terms undisclosed. [CR017, CR018, CR019, CR020, CR021, CR022]

7.4 People & Execution Risks

Founder Xu Jincheng serves simultaneously as legal representative, Chairman, and General Manager per Aiqicha registration data. This triple concentration of statutory liability, board governance, and operational management in one individual is acute even by early-stage deep-tech standards. Should Xu depart, become incapacitated, or face regulatory scrutiny in any jurisdiction, PaXini would face simultaneous governance, strategic, and execution disruptions. Co-founder Nie Xianru (8% stake, board director) brings commercial depth from ZTE, Xiaomi, and NetEase, but the public record does not document formal succession planning, independent board structures, or C-suite delegation of operational functions. PaXini's registered headcount is 315 insured employees per Aiqicha. This is modest relative to simultaneous demands: running five-region Chinese data collection factory operations, advancing OmniVTLA and HyperCosmos model development, manufacturing sensors and dexterous hands, assembling and deploying Tora humanoid robots, and preparing a public-company-grade organisation for an HK IPO within 12 to 18 months. The implied headcount gap between current capacity and IPO-ready operations is a material execution risk. The Waseda University founding provenance provides technical credibility but introduces an IP-provenance question: the extent to which foundational sensor and tactile intelligence IP was developed at the university before transfer versus post-founding has not been disclosed. No IP disputes with Waseda or the Sugano Lab have been identified, but absence of evidence is not confirmed clean provenance. International expansion ambition requires establishing non-Chinese legal entities and compliance infrastructure that the current team has not publicly addressed. [CR026, CR027, CR028, CR029, CR030, CR046]

7.5 Financial, Capital & Thesis-Break Triggers

PaXini is a pre-revenue unicorn. CBInsights records zero disclosed fiscal-year revenue across nine funding rounds through March 2026, with a post-Series-B valuation above RMB 10 billion. The closest functional comparable—Unitree Robotics, now STAR Market IPO-approved— reported RMB 1.699 billion in 2025 revenue with 60.13% gross margins. PaXini has published no analogous revenue milestone. Implied price-to-sales is incalculable; implied price-to-narrative is high. The planned Hong Kong IPO introduces bilateral risk. Favourable conditions would provide capital to fund the data factory expansion and OmniVTLA model training that management has explicitly cited as IPO-proceeds targets. Unfavourable HK market conditions—which have periodically closed the window for growth-stage deep-tech listings—would force a delay, bridge financing with further dilution, or a down-round. The compressed timeline between the March 2026 Series B close and a potential IPO application leaves limited runway to establish multi-deployment commercial proof before becoming a public company subject to quarterly scrutiny. Revenue model concentration carries structural risk. Reducing tactile sensor unit prices to as low as RMB 199 democratises adoption but requires significant volume to generate meaningful revenue in a global force sensor market growing at only 5% CAGR to USD 3.93 billion by 2035. Goldman Sachs projects the humanoid robotics market reaching USD 38 billion by 2035, providing a large-market narrative, but near-term realisation depends on production scaling that China's 140-plus manufacturers are also pursuing competitively. Goldman Sachs and Morgan Stanley research coverage positions PaXini favourably but reflects relationship-adjacent analysis rather than independent adversarial buy-side views. AgiBot reached 10,000 humanoid units in March 2026 via a listed-company acquisition route— an alternative IPO path signalling that direct HK Exchange registration carries approval risk for which PaXini may need contingency planning. Global competitors including Figure AI ($39B valuation, $1.9B raised) and Skild AI ($14B, $2B+ raised) command substantially larger capital pools and operate outside China's National Intelligence Law framework, giving them an inherent advantage with international enterprise and government customers. [CR031, CR032, CR033, CR034, CR035, CR036]

7.6 Exhibits

8.1 Financing Context and Valuation Anchors

PaXini completed its Series B financing on March 6, 2026, raising more than RMB 1 billion (~$144–150 million at prevailing exchange rates). The round was led by Huangpu River Capital, Kaitai Capital, and state-backed Xin'an Capital, with co-investment from Zhuhai Science and Technology Industry Group, Shanda Capital, Haichuan Fund, Guangzhou Emerging Fund, Uni-Trend, Zhilai Capital, LEO LION, Nanling Fund, and Xiangcheng Financial Holding. Existing investors including Addor Capital (Yida Capital) significantly increased their stakes. The post-money valuation exceeded RMB 10 billion (~$1.5 billion), making PaXini one of eight Chinese embodied intelligence companies to cross that threshold. Cumulative external capital raised stands at approximately $150 million across at least nine financing events since 2023. The RMB 10B valuation implies aggressive assumptions. No public revenue or gross-margin figures have been disclosed. CB Insights shows zero FY revenue for all tracked rounds. At an estimated run-rate revenue of $5–15 million (inferred from deployment scope and headcount), the implied revenue multiple spans 100–300x — a range only supportable if PaXini achieves rapid, scalable commercialization over the next 12–24 months. Goldman Sachs and Morgan Stanley research have named PaXini as an "upstream infrastructure layer" in embodied AI, with Goldman crediting it with more than 80% global tactile sensor market share. These analyst views support the category thesis but are not independently verifiable at this stage. A Bloomberg-reported Hong Kong IPO process — reportedly working with banks and potentially filing with HKEX in coming months — introduces an exit catalyst that, if realized, could resolve the valuation gap or crystallize it as a permanent overvaluation. [CV001, CV002, CV003, CV004, CV006, CV007]

8.2 Investment Thesis and Anti-Thesis

The investment thesis rests on four interlocking pillars. First, PaXini occupies the most differentiated position in the humanoid robot supply chain: high-precision tactile sensing. Tactile data is the one modality that cannot be scraped from the internet, and PaXini's ITPU architecture and DexH13 dexterous hand have already generated tens of billions of real-world multimodal data points, enabling closed-loop VTLA model training that no competitor currently replicates at scale. Second, the investor syndicate is an industrial validation signal. BYD, JD.com, TCL, SAIC, BAIC, and Meta-affiliate LEO LION constitute a supply-chain-to-deployment continuum. These are not passive financial investors; they are potential deployment customers and supply-chain anchors. Third, the Geely Qiantang deployment — eight continuous hours with 99.8% task success in automotive wiring harness assembly — provides the first credible evidence that PaXini's full-stack approach works in a demanding industrial environment. Fourth, the humanoid robot sector is entering a period of critical commercialization in H2 2026 per TrendForce, with Chinese output expected to grow 94% year-on-year; PaXini's upstream position makes it a supplier to the entire sector rather than a single-robot bet. The anti-thesis is equally pointed. Revenue is undisclosed, meaning the $1.5B valuation is priced entirely on narrative and pipeline, not on demonstrated unit economics. The Geely deployment is a single pilot; it does not establish recurring contract revenue. Unitree and AgiBot will capture approximately 80% of Chinese humanoid shipments in 2026 — and both are pursuing their own sensor and dexterous-hand integrations, threatening to internalize what PaXini currently provides. Goldman Sachs's >80% tactile market share figure is not independently verified; any competitor with cost parity would rapidly erode this share. Finally, as a Chinese company, PaXini is subject to Article 7 of China's National Intelligence Law and the broader geopolitical risks of Chinese AI hardware — risks that international enterprise customers and potential IPO investors on Western exchanges must weigh. [CV005, CV009, CV010, CV015, CV040, CV041]

8.3 Bull, Base, and Bear Scenarios

Three scenarios bracket the plausible valuation outcome. The bull case assumes PaXini becomes a de facto tactile standard for the humanoid robot ecosystem — effectively the Bosch Sensortec of embodied AI — with sensors and dexterous hands embedded in robot platforms from Unitree, AgiBot, GALBOT, and international OEMs. A successful HKEX listing in 2027 at a revenue-growth premium, combined with expanding data-mall licensing and OmniVTLA model commercialization, supports an exit valuation of $5–8 billion. Revenue in this scenario reaches $50–80 million by 2027 with high gross margins. The base case targets a 2028 IPO at $2–4 billion, assuming PaXini converts several pilot deployments into multi-year supply contracts, expands the tactile sensor installed base, and demonstrates recurring SaaS-style data licensing revenue. The bear case envisions slower-than-expected commercialization: pilots do not convert, Unitree accelerates its own sensor development, and a flattening Chinese robotics market compresses the TAM. A flat or down-round below $1.5 billion becomes necessary by late 2027 to attract growth capital, destroying early investor returns. [CV048, CV026, CV027, CV028, CV029, CV051]

8.4 Comparable Valuation Analysis

PaXini's $1.5 billion valuation sits in a broad peer range. Among US-based pure-play humanoid companies, Figure AI is the clear outlier at $39 billion post-Series C — reflecting both first-mover branding and BMW factory deployment receipts. Skild AI and Physical Intelligence command $14 billion and $11 billion respectively, but both are software/AI companies with near-zero capital equipment intensity. Apptronik's $5.5 billion valuation reflects a Series A stage with a $935 million raise and Mercedes-Benz as a strategic backer. NEURA Robotics at ~$4.6 billion is a better structural comp: hardware plus AI foundation model, European market, similar stage. Unitree's IPO target of $6.2 billion with disclosed 2025 revenue of RMB 1.699 billion and 60% gross margins is the most critical benchmark — it establishes what a commercially proven Chinese humanoid hardware company is worth on a public exchange. At $1.5 billion and likely $5–15 million in annual revenue, PaXini is priced at a substantial premium to Unitree's revenue multiple but at a steep discount to its US-listed peers. This spread is partly justified by PaXini's earlier stage and pre-revenue status, but it also reflects the Chinese market discount for geopolitical and regulatory risk relative to US comparables. The comparable set confirms that the sector supports high multiples for proven full-stack companies; PaXini's challenge is to close the proof gap. [CV016, CV017, CV018, CV019, CV020, CV021]

8.5 Exit Readiness, Diligence Asks, and Thesis-Break Triggers

PaXini's exit readiness is early but advancing. The Hong Kong IPO pathway is being actively explored per Bloomberg. Unitree's STAR Market listing committee approval on June 1, 2026 — the first embodied AI company cleared on a Chinese A-share exchange — has materially improved the IPO environment for sector peers. AgiBot is pursuing an alternative listing route. If Unitree's public offering completes at its $6.2 billion target, it will set a pricing floor for the entire sector and likely accelerate PaXini's HKEX filing. However, PaXini's IPO readiness depends on resolving key diligence gaps: audited revenue confirmation, customer contract terms, and cap-table disclosure. The most critical thesis-break triggers are failure to convert current pilots into recurring contracts within 18 months, loss of tactile market share to an integrated competitor, and a geopolitical export restriction targeting Chinese AI-enabled robotics hardware in Western markets. The legal dispute record (three business conflicts, one case filing, one court hearing on record at AiQiCha) warrants a clean-room review of any outstanding litigation before an IPO or anchor investment commitment. [CV030, CV031, CV032, CV033, CV034, CV035]

8.6 Exhibits