Psibot

1.1 Identity, product, and operating model

Psibot, also known as Lingchu Intelligence, is described as a Shanghai-founded embodied-AI company established in 2024. Its stated focus is dexterous manipulation rather than a consumer humanoid form factor. The company's product surfaces describe Psi SynRobot as a robot for logistics and industry, SynGlove Air as a real-machine teleoperation glove for data and control, and the PSI foundation-model family as the software layer. Third-party descriptions characterize Psi R1 as a vision-language-action model for long-horizon tasks, while the company's own pages present a broader product set spanning models, simulation, data, and robots. This supports a platform-plus-deployment thesis, not a cleanly disclosed recurring-revenue software model.[CO001, CO002, CO003]

1.2 Leadership, scale, and execution dependencies

Available reporting identifies Viktor Wang as a founder and chief executive, with prior JD.com and robotics experience, and names Xiaojie Chai, Yaodong Yang, and Yuanpei Chen among the founding team. The reported backgrounds span robotics, Peking University, and Stanford-linked AI research, which is relevant founder coverage for a technically demanding company. The evidence also points to logistics and fibre-optic manufacturing testing, a joint laboratory with Peking University, and a stated goal of collecting one million hours of data in 2026. These are operating signals rather than audited scale metrics: no source in the assigned pool discloses revenue, ARR, customer count, headcount, board composition, ownership percentages, or leadership turnover.[CO005, CO006, CO008, CO009]

1.3 Financing, strategic stakeholders, and milestones

Publicly available sources describe a March 2026 cumulative angel and Pre-A financing of approximately CNY 2 billion and a further August 2026 round of more than US$100 million. Other reporting describes the company as having raised about US$300 million since inception and places the August transaction at a US$1.48 billion valuation; the figures are not fully reconciled across sources and should not be treated as audited cash available. The investor mix is strategic as well as financial, including Chery-linked capital, Lens Technology, Tuopu Group, Wuhu investment entities, Fosun-related capital, and earlier state-backed and venture investors. The strategic rationale may improve industrial access while increasing related-party and concentration diligence needs.[CO010, CO011, CO012, CO013, CO014, CO015]

1.4 Exhibits

2.1 Market boundary and buying logic

PsiBot's relevant market is the software-and-data intelligence layer for robots used in logistics, manufacturing, retail, and other physical workflows, not the entire robot hardware stack. Its own site positions SynRobot for logistics, retail, and industry and describes a foundation-model-powered robot, while the refreshed funding coverage says the company develops VLA models rather than building complete robots. The included spend is therefore model licensing, deployment integration, teleoperation and data tooling, and potentially dexterous-hand or reference-system revenue; robot bodies, warehouse conveyors, generic industrial arms, and unrelated consumer robotics are excluded. Buyers are likely manufacturers, logistics operators, robot OEMs, and strategic component companies; users are operators and engineers, while the payer is usually an operations, automation, or innovation budget owner. The practical substitute is fixed automation or human labor in a constrained workflow, not another abstract AI model.[CM001, CM002, CM003, CM004]

2.2 Multiple sizing lenses and evidence limits

The available estimates establish a set of adjacent market lenses rather than a clean PsiBot SAM. GMI estimates the global humanoid-robot market at $10.9 billion in 2026 and $192.7 billion in 2035, with a 37.6% 2026–2035 CAGR; Grand View Research estimates global warehouse automation at $27.4 billion in 2026 and $59.5 billion in 2030, with an 18.7% 2024–2030 CAGR; and Fortune Business Insights estimates warehouse robotics at $7.35 billion in 2026 and $25.41 billion in 2034, with a 16.80% CAGR. These are not additive: warehouse robotics is nested within warehouse automation, while humanoid robots include hardware and applications beyond PsiBot's software layer. The 36Kr Research Institute gives a broader China embodied-intelligence estimate of RMB915 billion in 2025 and says it may exceed RMB1 trillion in 2026, but its boundary includes a broad ecosystem. IDC's deployment lens forecasts more than 510,000 humanoid shipments by 2030. The defensible conclusion is substantial adjacency growth, not a precise SAM or SOM.[CM005, CM006, CM007, CM008, CM009]

2.3 Drivers, constraints, and adoption path

Adoption is pulled by labor scarcity, e-commerce throughput, flexible manufacturing, and strategic industrial policy. The market sources repeatedly identify logistics, picking, packing, material handling, and manufacturing as early use cases where layouts and task flows can be bounded. IDC reports that more than 80% of users plan to deploy humanoids in palletizing, handling, picking, and machine tending over the next three years, while warehouse-market sources emphasize e-commerce and workforce shortages. China-specific policy is an additional demand catalyst: the 15th Five-Year Plan coverage describes embodied intelligence as a named future-industry priority and calls for training grounds, model evolution, core-component work, and humanoid deployment. Constraints remain material: published market analysis notes high initial investment, maintenance, safety, integration, and limited adaptation to unpredictable situations. PsiBot's industrial investors may accelerate pilots, but no assigned source verifies paid production deployments, pricing, recurring revenue, or unit economics.[CM010, CM011, CM012, CM013, CM014]

2.4 What the market means for underwriting

The market supports a focused beachhead thesis rather than an unrestricted platform assumption. PsiBot's logistics and advanced-manufacturing targets align with the most repeatable early workflows, and its reported dual-system model architecture and human operation data program are directionally relevant to deployment learning. However, the evidence is primarily market-level or company-originated and does not establish that PsiBot captures a defined share of any reported market. The most decision-relevant diligence is a bottom-up pipeline by workflow, robot OEM, geography, and deployment stage; contracted pricing and take rates; conversion from pilot to production; task-success, uptime, and integration metrics; and the share of demand that strategic investors can actually route. Until those data are available, the valuation should be stress-tested against a serviceable intelligence-layer market below the headline robotics estimates.[CM002, CM004, CM013, CM015]

2.5 Exhibits

3.1 The competitive boundary is wider than humanoid vendors

PsiBot competes for the intelligence layer of industrial and logistics automation rather than only for complete-robot sales. Third-party descriptions characterize Psi R1 as a vision-language-action model and PsiBot as a supplier of intelligence to manufacturers, while the company's own site presents a broader loop spanning foundation models, robot hardware, simulation, teleoperation, and data. This puts direct model vendors such as Physical Intelligence in the closest strategic peer set, but buyers can also choose a vertically integrated robot maker such as Unitree, AGIBOT, UBTECH, Figure, or Tesla; use NVIDIA's development stack; build an internal model; or retain fixed automation and conventional grippers. The alternatives therefore differ not only in capability but in ownership of hardware, deployment data, integration risk, and distribution.[CP001, CP002, CP006, CP012, CP014]

3.2 Architecture is differentiated, but proof remains issuer-led

PsiBot's strongest public differentiation is architectural. At a 2025 company event, it described cloud training followed by lightweight edge inference, real-world data collection through proprietary exoskeletons, supervised fine-tuning, and reinforcement learning. The same company presentation disclosed a meaningful performance gap against Figure's cited Helix frequencies: PsiBot reported 30-50 Hz for its low-level module and about 1 Hz for its high-level module, versus 200 Hz and 7-9 Hz respectively for Figure. Independent reporting also notes that PsiBot's L3 label, thirty-minute mahjong demonstration, and claimed data-cost advantage have not been independently verified. The evidence supports technical differentiation, not benchmark leadership or production reliability.[CP003, CP004, CP006, CP007]

3.3 Scaled hardware rivals possess stronger distribution evidence

Unitree and AGIBOT already present a different competitive asset: shipped hardware and manufacturing learning. Assigned sources report that Unitree generated RMB1.708 billion of 2025 revenue, shipped more than 5,500 humanoids, and had shifted more than half of its revenue mix toward humanoids by the first nine months of 2025. Yet another source reports that research and education represented 73.6% of Unitree's humanoid revenue from 2023 through the first three quarters of 2025, while industrial and smart-manufacturing uses represented 9%. That mix limits how confidently shipments can be equated with mature factory demand. For PsiBot, strategic investors such as Chery and Lens Technology may provide industrial access, but investment relationships are not evidence of arm's-length customer adoption or exclusive distribution.[CP005, CP008, CP009, CP010, CP011]

3.4 The moat depends on portability becoming a standard

A hardware-agnostic intelligence layer can diversify exposure across robot bodies, but it is durable only if customers prefer an external model over their own core software. That condition is not established. Independent coverage says major Chinese humanoid makers are developing VLA models internally, while U.S. model specialist Physical Intelligence is reportedly raising at an above-US$11 billion valuation despite an undefined commercialization timeline. Assigned sources also conflict on Physical Intelligence's commercial maturity: one tracker claims an API, more than thirty licensees, and US$500,000 annual fleet pricing, while another says commercialization timing is undefined. PsiBot's public list pricing, contract structure, retention, integration cycle, and hardware portability remain undisclosed in the retained evidence. Until those gaps close, switching costs and platform power should be treated as hypotheses rather than demonstrated moat.[CP013, CP015, CP016]

3.5 Exhibits

4.1 Monetization and revenue quality

PsiBot positions its intelligence layer for industrial, logistics, retail, and other embodied-AI deployments rather than presenting a conventional hardware product line. Its public surfaces present a logistics and industrial robot, a teleoperation glove, and a modular whole-machine solution, while third-party Psi R1 references describe a model intended for industry, logistics, retail, and end-customer applications. A separate industry account characterizes the strategy as licensing intelligence to manufacturers and pairing the model with a dexterous hand, simulation tooling, and training-data platforms. These descriptions establish plausible revenue mechanisms—model or platform licensing, integration or deployment work, and potentially hardware or data-system sales—but do not establish realized revenue, recurring revenue, take rate, backlog, or revenue mix. Official pricing, contract terms, discounting, and recognition policy are not publicly evidenced in the reviewed sources.[CI001, CI002, CI003, CI004]

4.2 GTM motion and unit economics

The available go-to-market signal is strategic and deployment-oriented: the July 2026 reporting says pilots were live with a major Chinese logistics vendor and a large fiber-optic cable maker, while the August briefing says funding will support research, human-operation data collection, and deployments in logistics and advanced manufacturing. PsiBot's own page presents logistics and industrial robot applications, a teleoperation glove, and task demonstrations, which is consistent with an integration-heavy enterprise motion. However, there is no public evidence here for sales-cycle length, customer acquisition cost, implementation fees, gross margin, service-level economics, utilization, payback, or expansion behavior. The likely commercial bottleneck is not only model performance but also deployment, integration, and data-collection cost per customer.[CI005, CI006, CI007, CI008]

4.3 Capital adequacy and financing dependency

PsiBot has access to substantial strategic financing signals, but financing raised is not cash available and a reported round is not a balance-sheet disclosure. A July 23 account attributed to Bloomberg described a nearly $100 million round at a $1.48 billion valuation and approximately $300 million of total funding since inception, while a separate August 28 briefing reported completion of a new round of over $100 million. The July account explicitly cautioned that its deal was close but not signed and terms could change, so the two reports should not be mechanically added. Neither assigned source discloses cash on hand, burn, runway, debt, working-capital needs, committed versus deployed capital, or the next financing trigger. The planned use of funds—research, data collection, and deployments—implies continuing capital needs before commercial scale is proven.[CI009, CI010, CI011, CI012]

4.4 Financial verdict and diligence blockers

The financial picture is investable only as an evidence-limited option on platform adoption, not as an underwritten cash-flow or margin story. PsiBot's model and data assets may support licensing, integration, and deployment revenue, but no assigned source supplies audited revenue, ARR, gross margin, customer concentration, contract duration, realized pricing, or cash runway. A competitor benchmark illustrates why those omissions matter: Unitree's 2025 prospectus materials report 2025 revenue of 1.70 billion yuan and gross margins around 60%, while a contemporaneous report describes first-quarter 2026 revenue growth alongside a sharp adjusted-profit decline and rising R&D and sales expense. That benchmark is not a PsiBot estimate; it shows the sector can scale revenue while profitability compresses. PsiBot should not be valued on revenue multiples until management provides a reconciled revenue bridge and evidence of repeatable customer economics.[CI013, CI014, CI015, CI016]

4.5 Exhibits

5.1 Product surface and customer workflow

PsiBot's customer-facing proposition is an intelligence layer for physical work rather than a single robot SKU. Its official English site describes Psi SynRobot as a robot for logistics and industry, SynGlove Air as a real-machine teleoperation glove for data collection and control, and multiple demonstrations spanning optical-module inspection, vacuuming, box packing, knot tying, object tossing, and Lego assembly. That surface maps to a workflow in which a buyer supplies a workcell and task, the platform acquires demonstrations or prompts, and the robot executes manipulation policies. The site also identifies partners, but does not disclose contract terms, deployment counts, uptime, or realized customer outcomes. A third-party model listing describes Psi R1 as a model rather than a physical robot and places it across industry, logistics, retail, and end-customer applications. Those are addressable use cases, not proof of commercial scale. The most defensible near-term wedge is therefore constrained logistics or industrial work where task boundaries and measurable cycle-time benefits can be specified before general-purpose autonomy is underwritten.[CE001, CE002, CE003, CE004, CE005]

5.2 Architecture, data loop, and maturity

The public technical description is consistent with a hierarchical vision-language-action design: a high-level model interprets language and visual context, while a lower-level policy converts representations and robot state into action chunks. The DexGraspVLA paper, authored by a Peking University/PsiBot joint-lab team and presented as an AAAI 2026 oral presentation, specifies a VLM-based planner, SAM and Cutie segmentation, DINOv2 visual encoders, a trainable ViT, projectors, and a diffusion transformer controller. It reports more than 90% grasping success across thousands of unseen cluttered scenes for that research system, using 2,094 successful episodes and roughly two hours of demonstrations. This is relevant technical evidence, but it is not a disclosed production benchmark for every PsiBot product. Company and trade coverage separately describe Psi R1 as hierarchical, end-to-end, reinforcement-learning based, and capable of a more than 30-minute Mahjong demonstration; those claims remain company-reported or derivative reports. The architecture's maturity signal is thus stronger at research-component level than at fleet-reliability level: buyers still need task-level acceptance tests, failure recovery measurements, data-governance terms, and integration evidence in their own environments.[CE006, CE007, CE008, CE009, CE010, CE011]

5.3 Differentiation, dependencies, and diligence blockers

PsiBot's potential differentiation is the combination of a model layer, dexterous manipulation, and a data-collection loop rather than a novel chassis alone. Origin of Bots attributes lower training-data collection cost to Psi-SynEngine and reports small-scale warehouse verification with improved sorting efficiency; both points are secondary-source summaries, not audited operating metrics. The academic paper offers a stronger mechanism-level signal for generalized grasping, while the official site shows breadth of task demonstrations. The counterweight is operational: the public materials do not specify model versioning, API or on-premise boundaries, observability, safety interlocks, incident history, service levels, or component/compute dependencies. A trade report also flags that independent verification of the Mahjong and data-cost demonstrations is scarce and discusses export-control and Chinese-law exposure for licensees. That report is not a legal opinion, and the absence of a public incident is not evidence that no incident occurred. Before deployment, diligence should require reproducible task benchmarks, network/data-flow diagrams, red-team and fail-safe results, support commitments, model-update controls, and a documented bill of materials and compute plan.[CE012, CE013, CE014]

5.4 Exhibits

6.1 Buyer segments and use cases

PsiBot is positioning its intelligence and dexterous-manipulation stack for industrial buyers rather than selling a single standardized consumer robot. Its official site describes products built for “Real-world Logistics, Retail and Industry,” while third-party descriptions place the model in industry, logistics, retail, e-commerce, and end-customer applications. The most economically legible starting jobs are repetitive but variable physical tasks: barcode scanning, sorting, kitting, packing, and fine manipulation. That segmentation is strategically coherent because buyers can evaluate seconds saved per pick or reduced manual handling, but the sources do not identify a paying buyer, contract value, deployment site, or production acceptance criterion. The customer thesis therefore rests on a plausible workflow fit, not yet on disclosed commercial conversion.[CU001, CU002, CU003, CU004]

6.2 Adoption evidence and durability

The evidence supports an adoption ladder, but not a revenue funnel. RoboAtlas records the company saying it has validated logistics-warehouse use cases at small scale and provides a business-contact channel; PsiBot's own materials show a logistics, retail, and industrial product surface. A separate article says the company cites deployments in semi-structured clothing warehouses for barcode scanning and sorting, while another report describes sector testing and partnerships as underway. These are materially stronger than a logo-only claim, but they remain company-linked descriptions without named sites, customer quotes, paid-versus-pilot status, utilization, uptime, repeat orders, or denominator. AInvest's sector analysis cautions that orders, pilots, and demonstrations are not the same as recurring demand. No source in the assigned pool discloses NRR, GRR, churn, renewal rate, contract length, customer count, or revenue; those are explicit gaps rather than zero values.[CU005, CU006, CU007, CU008, CU009]

6.3 Named proof, expansion, and concentration

PsiBot's strategic-capital base may create a route to early reference deployments, but it is not the same as independent customer proof. Chery led the reported 2026 financing and Lens Technology participated; the sources describe their industrial positions and potential supply-chain logic, yet do not establish that either company is a paying customer or that an investment converts into exclusive procurement. The available named-proof inventory is consequently partial: sources mention manufacturing, retail, e-commerce, logistics, and industrial settings, but do not name three contactable production customers. This creates two linked risks. First, strategic relationships could concentrate early demand around investors or related industrial channels. Second, the licensing thesis depends on repeatable integration across robot bodies, while the sources do not show renewal or multi-site expansion. The next diligence step is a customer-reference pack with legal counterparty, paid status, site count, production acceptance, outcome metric, renewal/expansion status, and related-party designation.[CU010, CU011, CU012, CU013, CU014]

6.4 Exhibits

7.1 Regulatory and legal exposure

China-specific cyber and data obligations are a material risk for an embodied-AI platform handling training data, robot telemetry, and potentially personal information. A legal analysis reports that amendments to China's Cybersecurity Law took effect on 2026-01-01, introduced AI governance language, broadened extraterritorial reach, and increased maximum fines to RMB10 million depending on the violation. A separate legal analysis likewise describes higher penalties, expanded reach, and the importance of documented remediation. These sources describe the legal framework, not a Psibot violation; the diligence question is whether its data flows, customers, and deployed systems trigger the relevant obligations. There is also no assigned primary source establishing Psibot's licenses, litigation history, or enforcement status. The investment implication is a potentially asymmetric downside: compliance controls can reduce exposure, but an undocumented cross-border or incident-response posture could delay deployments or create fines and remediation costs.[CR001, CR002, CR003, CR004, CR014]

7.2 Operational, quality, and technical execution

Psibot presents a full-stack model, data, simulation, and hardware strategy for logistics, retail, and industrial settings, but the strongest performance evidence in the assigned pool is either company-authored or a secondary summary of company demonstrations. The official site lists logistics and industrial products and model demonstrations, while third-party listings repeat the claimed Psi R1 long-horizon capabilities. Independent coverage explicitly says that Psibot's L3 classification and its demonstrations have scarce independent verification. That creates execution risk between a controlled demonstration and reliable production use: errors, integration time, safety controls, warranty exposure, and data-collection cost are not disclosed here. The broader robotics evidence also describes current deployments as limited and site-specific, and identifies dependence on foreign compute and precision-component inputs. The appropriate monitor is production evidence by site, task, uptime, safety incident, and unit economics rather than benchmark claims alone.[CR005, CR006, CR007, CR011, CR012, CR013]

7.3 Financial, partner, and concentration risk

The latest assigned reporting describes more than $100 million of new financing from industrial and state-linked investors, while separate coverage describes a nearly $100 million round at about a $1.48 billion valuation. Dealroom records a March 2026 $280 million financing at a $1.1 billion valuation and says the company targets high-value B2B scenarios, but it does not provide audited revenue, margins, burn, runway, customer concentration, or contract quality in the fetched text. Strategic investors can improve distribution and industrial access, yet they can also create related-party, channel, and customer-concentration questions. A platform strategy depends on manufacturers adopting the intelligence layer; if customers build models in-house or if a partner delays deployment, revenue conversion may lag the capital narrative. Until management provides cohort, contract, cash-use, and cap-table evidence, the financial risk should be treated as high-uncertainty rather than inferred from funding size.[CR008, CR009, CR010, CR015, CR016]

7.4 Exhibits

8.1 Price and evidence

The strongest current valuation evidence is a Bloomberg-reported financing described by the Strait Times as close to US$100 million at a US$1.48 billion valuation, with about US$300 million raised since inception. Yicai independently reports total capital of CNY2 billion (US$300 million), while Gasgoo reports a new funding round used for R&D and commercialization without publishing a valuation. These sources establish a financing context, not a verified revenue multiple or cash balance. The round's strategic investor mix is economically relevant: Humanoid Intel reports participation by Ningbo Tuopu, Lens Technology, a Chery-backed fund, Wuhu Investment Holding, Fosun Fortune Capital, 37 Interactive Entertainment, and existing investor Zhuhai Technology Industry Group. Strategic capital may improve access to deployment sites, but it can also encode industrial-policy or supply-chain objectives that are not equivalent to financial underwriting.[CV001, CV002, CV003, CV004, CV005]

8.2 Scenarios and comparables

The available comparable evidence is a capital-market and funding landscape, not a like-for-like public-company valuation set. AI Funding Tracker reports large 2026 rounds for model-layer and humanoid companies, including Skild AI at more than $14 billion and Physical Intelligence at $5.6 billion in its cited set, while Yicai reports that Chinese robotics companies above CNY10 billion in valuation raised more than CNY40 billion in the first half of 2026. Those benchmarks show that the sector can support high software-layer prices, but they do not establish Psibot's relative multiple because Psibot has not publicly disclosed revenue, recurring revenue, gross margin, or shipment economics. ChinaBizInsider's account of Unitree, DEEP Robotics, and Leju shows how IPO benchmarks can separate profitable, subsidized, and loss-making businesses; the same discipline should be applied to Psibot before treating a headline valuation as fair.[CV006, CV007, CV008, CV009, CV010]

8.3 Decision and diligence

The investable thesis is that Psibot can become an intelligence-layer supplier whose models and data systems scale across industrial customers. That thesis is not yet demonstrably asset-light: the official site also markets Psi SynRobot for logistics and industry plus teleoperation hardware, so diligence must separate licensable software economics from hardware and integration capital needs. RobotAIGeek reports a fifteen-fold valuation increase in roughly a year and describes the company's own L3 and data-cost claims as independently unverified; its account also flags dependence on foreign compute and component foundations. The appropriate decision is research-more at the reported mark: require audited financials, cap table and preference terms, named customer references, deployment uptime and gross-margin evidence, and independent validation of model demonstrations. A move toward track or buy would require evidence that recurring commercial revenue and reliability justify the premium; a down-round, failed deployment, or inability to verify customer economics would break the thesis.[CV011, CV012, CV013, CV014, CV015, CV016]

8.4 Exhibits