Legal name: Sudo Tech
Sudo Tech Deep Dive: Zero-Real-Data Simulation Training, the Fully In-House Sudo R1 Stack, and an Atomic-Skill Commercialisation Strategy from a Startup Valued Above CNY 20 Billion in 17 Months
Sudo Tech trains robots entirely in simulation with no real-robot data, ships the fully in-house Sudo R1 stack, claims above 99 percent atomic-skill success rates, and has completed a proof of concept at a battery-module assembly line; cumulative funding exceeds CNY 5 billion at a valuation above CNY 20 billion.
Executive Summary
Sudo Tech (Shanghai Sudo Technology Co., Ltd.) is a general embodied intelligence company built around a single unconventional bet: train robots entirely in simulation, with no real-robot demonstration data in the training loop. The company was registered in Shanghai's Yangpu district on May 19, 2025, is wholly owned through Sudo Holdings HK Limited, and is led by co-founder and CEO Han Zheng alongside co-founder, chairman and CTO Su Hao, who authored ImageNet, led the creation of ShapeNet, co-created PointNet, built the SAPIEN simulator and the ManiSkill benchmark, and developed the TD-MPC model-based reinforcement learning algorithm. The founding pair are former Beihang classmates with more than twenty years of shared history, and Han Zheng is on his third venture after ZEPP and Rocket Technology, both acquired.
The technology centrepiece is Sudo R1, released on April 20, 2026, a fully in-house robot system combining a 3D world model with reinforcement learning in one architecture. The company reports a 98 percent first-attempt grasp success rate on previously unseen objects, close to 100 percent within two attempts, and 99.5 percent on industrial manipulation, all without any real-robot training data, with the simulation engine producing more than a million training samples per day. Hardware is also in-house: a seven-degree-of-freedom arm per side, a 22-degree-of-freedom directly driven dexterous hand, and a wheeled dual-arm platform, with a ten-thousand-unit production base filed in Shanghai Lingang.
Commercialisation is deliberately narrow. The flagship reference is a battery-module assembly line built with CATL, where four robots cooperated through loading, cell assembly, dust removal and scanning, and finished down-feed, with adaptation to a second battery plant taking roughly one to two days. Han Zheng has publicly placed the company in the second of three commercialisation stages, with no scaled paying deployment disclosed, and the plan is proof of concept plus small batches in late 2026, hundreds of units in 2027 and thousands in 2028.
Funding is the strongest part of the story: a 500 million US dollar Pre-A round in April 2026 at a valuation above 2 billion US dollars, followed by a new round above CNY 1 billion led by Shanghai state-backed investors and Guotai Haitong in September 2026, with cumulative funding reported above CNY 5 billion across five rounds in 17 months. The central open question for a technical diligence is whether laboratory success rates transfer to a dirty, reflective, tolerance-driven factory floor, and whether a company with roughly 250 people can execute its stated path to thousands of units per year.
Company Overview
Sudo Tech was incorporated as Shanghai Sudo Technology Co., Ltd. on May 19, 2025, with a unified social credit code of 91310110MAEJEQCP87 and Han Zheng as legal representative. The registered address is on Yangshupu Road in Yangpu, the registered capital was raised from CNY 10 million to CNY 100 million on October 17, 2025 and to CNY 200 million on March 28, 2026, with CNY 100 million paid in, and the enterprise type is a limited liability company wholly owned by a Hong Kong entity, Sudo Holdings HK Limited. A Shenzhen branch was added in July 2026 under Han Zheng, and a wholly owned Shanghai Sudo Robotics was incorporated in August 2026. The registered business scope covers other technical promotion services, with import and export added in June 2026. The name is drawn from the Unix command: an ordinary request is refused, while the same request prefixed with sudo is executed.
The overseas structure is a separate but related entity. Hillbot was founded in the United States in 2024 by Su Hao with Han Zheng joining as co-founder, and is commonly described as a twin body holding the academic assets such as SAPIEN, while Sudo Tech carries the domestic commercialisation and the hardware platform. Core team members include long-standing collaborators from Hillbot, and Han Zheng has said the core team took close to ten years to assemble. Su Hao returned to China full-time in 2026 while holding a professorship at Fudan University and the role of founding dean of the Institute for General Physical Intelligence.
The leadership team covers five distinct capability gaps. Su Hao sets the technical agenda. Han Zheng handles operations and commercialisation, having founded the motion-control hardware company Zekep in 2009, co-founded ZEPP in 2010 which was acquired by Huami in 2018, and founded the smart-office platform Rocket Technology in 2017 which was acquired by Ucommune in September 2018. Zhang Xiaoheng, who joined in December 2025 as chief strategy officer, previously worked at ABB and Huawei and later invested at BlueRun Ventures. Chen Runze, who joined as hardware lead, came from Source Code Capital as an investor and had previously backed Unitree. Xu Zexiang, responsible for large-model research, previously led generative-AI work at Adobe and has more than 11,000 Google Scholar citations. Martina Hansen serves as co-founder and chief operating officer, with Columbia and Harvard Business School backgrounds and prior roles at Covariant and Verity. Dai Xiaowei is also listed as a co-founder, and Xiao Jinrong is finance lead.
The composition of the team is an operational fact rather than a branding choice. Han Zheng has stated that roughly half of the organisation sits in the brain, that is models and algorithms, around three tenths in hardware, and two tenths in commercial operations, and that the headcount stands near 250. He also notes that a significant share of core research staff are fresh graduates, a deliberate choice to cultivate talent from source.
Technology
Sudo Tech describes its core path as Real2Sim2Real, with synthetic data and simulation carrying the bulk of training and a small amount of real-robot data used only to correct systematic simulator-to-reality bias. The company positions this as a cost argument: top-tier simulators drive the marginal cost of data acquisition down sharply, which fits the first principles of scaling, whereas teleoperation collection is expensive, slow, and poorly suited to the long tail of unknown objects and unusual working conditions. The in-house simulation engine is designed to reproduce friction, collision and dynamic constraints rather than merely rendering plausible frames, and it can produce more than a million training samples per day. The company describes itself as one of the few Chinese firms that builds a native embodied simulation engine in-house and competes globally in that layer.
The architectural commitment is world model and reinforcement learning in one design rather than a pipeline of separate perception, planning and control modules. Sudo R1 conditions each control step on the robot's latest observation and adapts actions accordingly, running a closed-loop policy at 15 to 25 hertz, which the company contrasts with action chunking approaches that effectively observe once per second. The system reasons about 3D obstacles and available free space, and can bypass obstructions and avoid collisions rather than treating the workspace as a flat plane. The company also points to a layered split, with an upper model responsible for instruction understanding, environment analysis and task decomposition, and lower-level skills responsible for reliably executing grasps and insertions, on the argument that the upper model may be general while contact-rich skills must be dependable.
The unit of capability management is the atomic skill. The company named the reusable building block the atomic capability, covering precise placement, flexible-object manipulation, two-handed coordination, mobile grasping and autonomous navigation. The commercial priority is a small set of high-frequency primitives, grasp, place, insert and fold, brought above 99 percent success and then packaged as callable software modules. Han Zheng has framed generalisation as secondary: the premise of scaled deployment is high reliability, and a laboratory that occasionally succeeds and then patches it with more data is still far from commercial. An investor has told QuantumBit that the team moved from one skill to more than ten in three months, attributing this to reuse, combination and migration rather than retraining from scratch.
On hardware, the company uses a wheeled base with dual arms and a two-finger gripper in the near term, arguing that bipedal platforms divert compute and engineering effort into balance control while the operative value lies in the hands. It has developed a seven-degree-of-freedom arm per side and a 22-degree-of-freedom directly driven dexterous hand, which is unusual among Chinese embodied startups, and is planning both an industrial heavy-duty and a general commercial light-duty variant. Han Zheng describes self-developing the body as forced rather than strategic: from 2022 the team tested various mobile bases and whole-body robots whose architectures and interfaces could not meet the algorithm requirements, and attempts to commission custom hardware from third parties failed because specialised suppliers cannot integrate whole-machine design, controller development and low-level interfaces in reasonable time.
The clearest statement of the company's own standard is Su Hao's observation that the breakthrough in physical intelligence lies not in model architecture but in the aggregation of knowledge, and that the industry's centre of gravity should shift from how impressive a demonstration is to how reliable the operation is. He has also warned that many excellent demonstrations still sit across a decisive gap from genuine general capability, and in a Fudan lecture he separated three confusions that do not constitute embodied intelligence: a large model bolted to a robotic arm, a humanoid robot, and AI-generated video.
Product and Commercial Model
Sudo R1 was released on April 20, 2026, together with a sixty-minute unedited continuous evaluation video. The robot faced more than a hundred previously unseen objects including transparent cups, reflective metal parts, soft cloth and irregular items, under changing illumination, dynamic backgrounds such as a television screen, physical interference and obstacle constraints. The reported first-attempt success rate was about 98 percent, close to 100 percent within two attempts, with roughly two hundred grasps in the run and no cherry-picking. At the World Artificial Intelligence Conference in July 2026 the capability set expanded to more than ten skills including bimanual operation, flexible-object handling, precision assembly, insertion, parcel wrapping and autonomous navigation, and the team also demonstrated an end-to-end whole-body control policy for mobile manipulation in which perception, base, torso and arms are not split into separate modules.
The industrial reference is the battery-module line built with CATL, first shown at WAIC 2026 and later brought to the Yunqi Conference in September. Four wheeled dual-arm robots cooperated through cell loading, cell assembly, dust removal and barcode scanning, and finished down-feed, with a reported cycle under three minutes and around fifteen minutes for ten sets of twenty cells. Media timing at the show found the pace still slower than human operators, which the company's software engineering lead acknowledged directly, saying the next step is to use a capable brain to extract the full potential of the hardware. The collaboration started from a cable precision-insertion station and uses fused vision and tactile sensing to address contact assembly that pure vision cannot resolve, with the first process station having completed proof of concept.
The commercial model has three layers. The first is the atomic skill library, licensed or called by integrators who assemble robots themselves for battery assembly, automotive, retail and other structured manufacturing. The second is platform and body sales, where Sudo sells both the hardware and the operating system and positions itself as an embodied developer platform in the shape of the iPhone and iOS pairing, with an SDK so partners build applications on top. The third is developer ecosystem services, including deployment support in more than a dozen planned developer centres across China, the United States and Europe.
The rollout plan is explicit: proof of concept and small-batch delivery with leading customers in the second half of 2026, hundred-unit deliveries in 2027 and thousand-unit scale in 2028, with a ten-thousand-unit production base already filed in Shanghai Lingang. The company targets serving one thousand industry developers in the hope that five to ten of them build applications with real scale, a formula Han Zheng has used to describe what he regards as the iPhone moment. Han Zheng has also been unusually direct about current stage, describing commercialisation as being in its second stage, the proof-of-concept phase, with no scaled paying deployment revenue disclosed, and Zhang Xiaoheng has argued that investors must accept the long cycle of technical iteration rather than press for short-term orders.
Competitive Landscape
Han Zheng's own segmentation of the field is useful because it is specific. He places Physical Intelligence and Skild AI in the camp that builds a brain and stays hardware-agnostic, Figure in the body-plus-model camp whose model side is, in his words, somewhat weaker and less focused, and Tesla Optimus as similar in philosophy to Sudo by opening the hardware to third-party developers while not yet providing a low-level manipulation model. The combination he actually watches is DeepMind together with Boston Dynamics, and that judgement is itself an argument for treating the simulator as a core asset rather than a tool.
Against Chinese peers, the differentiation is the ownership of the simulation layer. SAPIEN and ManiSkill are the academic ancestors of the current stack and are used worldwide by groups studying robot learning in simulation, which means the foundational assets were validated by the research community before commercial validation. The competitive set in China, including the foundation-model startups, the data-infrastructure players and the model-plus-body players, has converged on a similar claim about general purpose physical intelligence; Sudo's version of that claim is unusual in one respect, that training data comes entirely from simulation and that the company asserts the resulting policy transfers without task-specific real-robot data.
On the industrial side, the comparison that matters is against incumbent automation. The commercial argument is not peak performance but migration cost. In the traditional model, adapting a robot to a new station requires custom development and a stoppage for data collection, with material consumed in the process; Sudo reports that an integrator familiar with the system needs roughly one to two days to deploy at a different battery plant by calling a public skill library, with process logic re-composed into a new workflow. Han Zheng notes that most potential customers cannot open their lines for data collection, which is precisely why the simulator-first route is commercially relevant rather than merely elegant.
The tension in the competitive picture is that the same asset which differentiates Sudo also narrows its market. A policy that never saw a factory will face oil, dust, dimensional tolerance, mirror reflection and unmodelled contact. Han Zheng has himself stated that rigid grasping is comparatively easy while complex contact-rich operations still require real information, which narrows the industrial claim to the subset of tasks where rigid grasping and structured insertion dominate.
Financing and Capital Structure
The capital structure has assembled quickly and from unusually varied sources. An angel round in 2024 with Tsinghua-era seed investors, and a seed round in 2025 with Tusi Star Venture Capital, precede a Pre-A++ round in January 2026 led by China Life Equity, at which point the company was only months old, had not yet released a product and had no visible commercialisation path. On April 20, 2026 Sudo announced a 500 million US dollar Pre-A round, roughly CNY 3.41 billion, at a valuation above 2 billion US dollars, with participation from CATL, Alibaba, Tencent, Ant Group, Hillhouse, IDG, BlueRun, Fortune Capital, Fudan Science and Technology Venture Capital, Yunhui Capital, Green Oasis Capital, China Life Equity and Legend Capital, with Gaohe Capital acting as long-term exclusive financial adviser. On September 30, 2026 the company announced a further round above CNY 1 billion led jointly by Shanghai Guotou Xiandao, Guotai Haitong and Fortune Capital, with Daohe Longterm, Dongfeng Asset, CDB Financial, Longma Peak and Liangtu Zhice participating and Xiaomi and other existing shareholders following on. Cumulative funding is reported above CNY 5 billion across five rounds in seventeen months.
The shareholder register is as informative as the amounts. Industrial capital includes CATL, which is both an investor and a co-development partner, plus Xiaomi, NIO, Alibaba, Tencent and Ant Group; state and long-duration capital includes China Life Equity, Shanghai Guotou Xiandao and CDB Financial; and the financial investors include Hillhouse, IDG, BlueRun, Legend and Green Oasis. The presence of state-backed insurance money in a company that reported no scaled revenue is the single most striking capital-structure fact, because it signals that a long-horizon strategic assessment of physical intelligence is being funded well ahead of commercial proof.
The valuation has two circulating numbers and they are not interchangeable. One source set reports a post-money valuation above 2 billion US dollars, roughly CNY 13.4 to 13.6 billion at prevailing rates; another reports CNY 20 billion. The gap is too large to be an exchange-rate artefact and is best understood as two different measurement points, one anchored to the April round and the other a later media estimate. The same caution applies to the cumulative funding figure, which is a media aggregate rather than a company-disclosed total.
DISCREPANCIES
1. The title of Su Hao. Sources differ on whether the co-founder and chief technical officer of the company is Su Hao, who is described as chairman and chief technology officer in company materials, founder and chief technology officer in database records, and chief technical adviser in some media coverage. Source grade: OFFICIAL versus MEDIA.
2. The founding date. The company entity was registered on May 19, 2025, the overseas sibling Hillbot was founded in 2024, and Han Zheng has described commercial preparation beginning in 2022. Source grade: OFFICIAL versus MEDIA.
3. The sequence of round names. A Pre-A++ round led by China Life Equity is reported for January 2026 and a Pre-A round of 500 million US dollars for April 2026, which places the higher-suffix round before the base round in time. Source grade: MEDIA.
4. The valuation figure. Post-money valuation is reported both as above 2 billion US dollars and as above CNY 20 billion, which at prevailing exchange rates differ by roughly forty percent. Source grade: MEDIA, two incompatible figures.
5. Headcount. The registered insured employee count is 42, the media figure is around 50 to 200, and Han Zheng states the organisation is around 250 people with a five-three-two split across brain, hardware and commercial functions. Source grade: OFFICIAL versus MEDIA.
6. The grasp success rate. Four figures circulate: 98 percent first-attempt on unseen objects, close to 100 percent within two attempts, 99.5 percent on industrial manipulation, and a company statement of above 99 percent as the precondition for commercialisation, each referring to a different environment. Source grade: OFFICIAL, self-reported.
7. The depth of the CATL relationship. The battery-module line has been described variously as a joint development, as a four-robot demonstration at a public exhibition, and as a completed proof of concept at the first process station. Source grade: OFFICIAL versus MEDIA.
8. Cumulative funding. The aggregate is reported as above CNY 5 billion across five rounds, while the disclosed amounts across the individual rounds do not obviously sum to that figure. Source grade: MEDIA.
Risks and Diligence Gaps
The central diligence gap is the transfer of laboratory performance to a production floor. The 98 percent figure comes from a test bench under controlled lighting and a controlled background, whereas factory conditions introduce oil, dust, dimensional tolerance and mirror-like reflection. The industry response, quoted in Chinese manufacturing press, is that the questions are not about whether grasping works but about the boundary conditions of the metric, the reproducibility of the test and the size of the simulator-to-reality gap, and that no independent third-party replication exists. Su Hao himself has said that many fine demonstrations remain separated from general capability by a key gap.
A second gap sits in the missing denominators. There is no disclosure of revenue, of order backlog, of unit shipments, of gross margin, of burn rate or of headcount trajectory against plan. The company describes itself as being in the proof-of-concept stage of commercialisation, and the disclosed reference customers are largely unnamed for retail, catering and logistics. Any statement about commercial traction currently rests on demonstration footage and a deployment roadmap rather than on audited operating data.
A third gap concerns the reliability of long-horizon tasks. Long-sequence success is approximately the product of per-step success rates, so a primitive that needs occasional adjustment compounds quickly over a multi-stage workflow. The company has demonstrated four-robot division of labour rather than a single unit completing a long sequence, and the new skills shown in July 2026 were not accompanied by the uniform per-skill success figures given for grasping. Whether the reuse-and-composition story holds at higher task lengths is not yet observable.
A fourth gap is organisational. The stated path from roughly 250 people to a thousand-person delivery organisation, spanning developer centres across three regions and a ten-thousand-unit plant, has no disclosed staffing plan, and technical direction is described as depending on Su Hao and a small number of core technical leads. There is also a related concentration risk in the founder structure, since the Chinese embodied sector remains populated by teams drawn from the same academic lineage.
A fifth gap is the valuation-versus-delivery gap. A CNY 20 billion valuation with no scaled paying revenue is a bet on the platform narrative rather than on current operations, and the company has chosen the harder path explicitly, refusing vertical customisation before the underlying paradigm is proven. That choice is defensible on a decade view and expensive on a funding-cycle view.
Finally, the licensing and intellectual property position of the simulation stack inherited from the academic line has not been made public, and the relationship between the overseas entity and the domestic operating company is described in the sources in narrative rather than documented terms. These are diligence items rather than red flags, and they are addressable with primary documentation.
Overall Assessment
Technology scores 8.5. The academic provenance of the stack is the strongest single asset on this list, since ImageNet, ShapeNet, PointNet, SAPIEN, ManiSkill and TD-MPC are all contributions of the same founder, and a simulator built in-house with more than a million samples per day is a hard capability to replicate. The integration of world model and reinforcement learning with a closed-loop policy at 15 to 25 hertz is technically coherent, and the argument for pushing atomic primitives above 99 percent before composing them is the right engineering order. Deducted points: no third-party benchmark, no published simulator-to-reality gap study, and the 98 percent figure rests on a self-run video.
Commercialisation scores 6.5. The CATL line is the most credible industrial reference in the Chinese embodied sector and the one-to-two-day adaptation cycle is a genuinely differentiated metric against incumbent custom development, while a filed ten-thousand-unit plant and a named plan for developer centres show real preparation. Deducted points: no disclosed revenue or backlog, the demonstration pace still slower than human operators, the company's own statement that it sits at the proof-of-concept stage, and retail, catering and logistics references that remain unnamed.
Financing scores 9.0. A 500 million US dollar Pre-A at a valuation above 2 billion US dollars, a further round above CNY 1 billion four months later led by state-backed investors and a broker, cumulative funding reported above CNY 5 billion in seventeen months, and a shareholder register combining CATL, Alibaba, Tencent, Ant Group, Xiaomi and NIO with China Life Equity, CDB Financial and the major financial institutions is among the strongest records in the sector. Deducted points: valuation stated on two incompatible bases, cumulative funding a media aggregate, and state capital entering ahead of any scaled revenue.
Product maturity scores 7.0. A full-stack system with in-house arms, a 22-degree-of-freedom directly driven hand, a unified simulation and learning architecture and a documented skill library is a real product rather than a demonstration, and the skill-module licensing model is workable for integrators who already build robot cells. Deducted points: the industrial deployment is a single station at proof-of-concept stage, no unit shipment figures, and the developer platform ambitions rest on a thousand-developer target rather than an existing ecosystem.
Market position scores 8.0. Sudo sits at the centre of the Chinese general embodied intelligence cluster with a valuation among the highest and one of the few simulation-native stacks, and the industrial capital in the register suggests it is treated as a platform asset rather than a product vendor. Deducted points: the field is crowded and funding has drawn competitors to the same simulator-first argument, the iPhone framing invites the same comparison that ended unfavourably for other platform analogies, and regulatory and geopolitical exposure applies to a company with a Hong Kong holding structure and United States research heritage.
The composite view is 7.8. Sudo Tech is the clearest case in this list of a company whose central technical bet was formed in academic work two decades before it became a company, and the evidence that the bet is real is stronger than the evidence that it will pay off on schedule. The intellectual contribution, a simulator-native training pipeline that refuses to depend on real-robot data, is genuine and the industrial reference with CATL is the most persuasive field evidence in the Chinese embodied sector so far. What is missing is the part that decides outcomes, which is operating data: audited revenue, unit shipments, line-side failure rates over weeks rather than minutes, and evidence that the developer ecosystem arrives rather than being recruited. For physical AI as an investment and adoption theme, the significance of this company is larger than its current commercial weight, and the most informative next disclosures will be the 2027 hundred-unit delivery figure and a published simulator-to-realty gap study.
References
- Sudo Tech official site, sudo.ai, accessed 2026: Sudo R1 release and sixty-minute unedited evaluation video, world model and reinforcement learning integrated design, closed-loop policy at 15 to 25 hertz, obstacle reasoning, atomic capability list, developer platform positioning.
- Baidu Baike entry for Shanghai Sudo Technology Co., Ltd., accessed 2026: registration on May 19, 2025, unified social credit code 91310110MAEJEQCP87, registered capital progression to CNY 200 million with CNY 100 million paid in, wholly owned by Sudo Holdings HK Limited, Shenzhen branch in July 2026, wholly owned Shanghai Sudo Robotics in August 2026, share and trademark registrations between December 2025 and April 2026, named officers including Su Hao, Han Zheng, Dai Xiaowei, Xu Zexiang, Chen Runze, Zhang Xiaoheng and Xiao Jinrong.
- Tencent News, September 30, 2026, report on the new round above CNY 1 billion led by Shanghai Guotou Xiandao, Guotai Haitong and Fortune Capital, with Daohe Longterm, Dongfeng Asset, CDB Financial, Longma Peak and Liangtu Zhice participating and Xiaomi following on.
- Sina Finance, October 14, 2026, feature on the two Beihang alumni and the simulation bet, covering the April 20, 2026 Sudo R1 release, the 98 percent and 99.5 percent figures, the three-layer skill architecture, Han Zheng on reliability preceding generalisation, the WAIC battery-module line timed at under three minutes and roughly fifteen minutes for ten sets of twenty cells, one-to-two-day migration to a second battery plant, the 2026 to 2028 delivery plan and the Lingang ten-thousand-unit base.
- QuantumBit interview with Han Zheng at WAIC 2026, July 2026: the extreme-reliability argument, the difficulty of line stoppage for data collection, internal debugging within one day for a new cell model, the public skill library for integrators, the forced rationale for in-house hardware development, and the iPhone plus iOS long-term platform framing.
- iitime, feature on Sudo R1 entering the factory, 2026: the controversy over metric definition, test boundaries and reproducibility, the absence of independent third-party replication, the distinction between the laboratory benchmark and industrial deployment, and the simulation engine generating million-scale samples per day.
- Toutiao, feature on the new round and the simulation bet, September 2026: the company name origin, the nearly ten-year team assembly, the 250-person figure with a five-three-two split, the in-house wheeled dual-arm platform with a seven-degree-of-freedom arm and a 22-degree-of-freedom directly driven hand, the WAIC ten-plus skills, the CATL collaboration rationale, the 2026 to 2027 commercial plan and the near ten-thousand-card compute cluster.
- TMTHome, report on the new round above CNY 1 billion, 2026: the atomic skill priority on grasp, place, insert and fold above 99 percent, the training loop of synthetic data plus simulation plus limited real-robot data, the developer and integrator ecosystem model, the Pre-A round structure and the rationale of China Life Equity as long-duration capital.
- Tencent News, August 25, 2026, coverage of Han Zheng at the 2026 World Robot Conference: the statement that scaled deployment means high-reliability generalisation on a limited set of valuable skills before skill composition and orchestration, and the observation that research benchmarks remain far from commercial standards.
- Lao Hu Cai Jing, September 2026: the funding timeline including the January 2026 Pre-A++ round led by China Life Equity, the April 20, 2026 Pre-A round of 500 million US dollars at a valuation above 2 billion US dollars with Gaohe Capital as long-term exclusive financial adviser, and cumulative funding above CNY 5 billion.
- TMTech and TechMorse feature on Sudo Tech, 2026: the iPhone moment framing from the WAIC main forum, the distinguishing role of knowledge aggregation in physical intelligence, the three confusions that do not constitute embodied intelligence as listed in a Fudan lecture, and the description of the platform as hardware plus operating system plus developer ecosystem.
- Weikehao product analysis of the Sudo R1 evaluation, 2026: the layered architecture, the atomic capability framing, the reported grasp success figures under changing illumination and dynamic backgrounds, the flexible-object and insertion demonstrations, and the observation that new skills were shown without uniform per-skill quantification.
This report is compiled from public sources and does not constitute investment advice. The company is private with no audited financials; the 98 percent grasp figure, the 99.5 percent industrial figure, the more than one million daily simulation samples, the two hundred and fifty headcount and the deployment roadmap are company statements without third-party replication, the valuation appears both as above 2 billion US dollars and as above CNY 20 billion, cumulative funding is a media aggregate rather than a disclosed total, and the CATL relationship has been described at three different levels of depth; uncertainties are flagged throughout. Data is current as of 5 November 2026.
Related papers
- AffordanceWAM: Affordance-Aware Joint World-Action Modeling for Robot Manipulation
- NeuIDO: Neural Intrinsic Dynamics Operator for Physics-Informed 4D World Models
- ForceTwin: Physics-informed Digital Twins for Robotic Manipulation from Instrumented Human Interaction
- CHOREO: Every Humanoid Skill as a Trajectory
- HOTICE: Whole-Body Humanoid Object Transportation in Cluttered Environments
Related tools
Related cases
Related reports66
- 51World Deep Dive: A Listed Digital Twin Company Where 51Sim Becomes the First Growth Engine While 51Earth Nearly Vanishes, and Two Irreconcilable Financing Naming Conventions Must Both Be Kept
- AMI Labs: A USD 1.03B Seed Bet on World Models, LeCun's Non-Generative Path and Its Blank Scorecard
- Anshi Pacific (PERA) Deep Dive: A Beijing Domestic-CAE Leader Founded in 2003 Whose PERA SIM General Simulation, AIROM Reduced-Order Models, AgentX Multi-Agent and Jingzhi iGPT/Whale Span from Industrial Simulation to Physics AI
- AspenTech Deep Dive: The Process-Simulation Incumbent Acquired by Emerson and Its Hybrid Industrial-AI Route
- Autodesk Deep Dive: Neural CAD Foundation Models and a $200M World Labs Bet on Physical-World AI, with NAVASTO Adding CFD Surrogate Capability
- Bentley Systems Deep Dive: iTwin Digital Twin and Generative AI Empower Full-Lifecycle Infrastructure Engineering
- BeyondMath Deep Dive: A Foundation Physics AI on the Neural-Operator Route, a $19M Consortium Budget, and Three Misread Facts
- Cadence Deep Dive: System-Level Multiphysics EDA and the Millennium M2000 Supercomputer Anchor Physical AI, with AuraStack Agentic AI and Fidelity GPU CFD as Key Levers
- COMSOL Deep Dive: Four Decades of Multiphysics Solvers, Entering Physical AI via DNN Surrogate-Model Training and All-Physics GPU Acceleration
- Dassault Systèmes Deep Dive: 3DEXPERIENCE and SIMULIA Virtual Twin Physics Behavior Anchor Industrial AI, with LEO Virtual Companion and GPU Explicit Dynamics as Physical-AI Levers
- DP Technology: From the Gordon Bell Prize to National-Level Unicorn Status and the Door of a Hong Kong Listing, the AI-for-Science Leader's Technical Delivery and Commercial Blank
- Emmi AI: From Linz Spin-out to a EUR 330M Mistral Acquisition in 17 Months
- Fainite Deep Dive: A Zurich Physics-AI Simulation Startup Whose Caltech, ETH Zurich, Cambridge and Google Team Uses Physics-Aware Deep Learning to Cut Simulation from Tens of Hours to Seconds, With Only Venture Kick Grants and No Priced Round
- Fysics AI: The PhysX Founder's Second Act, a Domestic Differentiable Physics Engine and Its Ecosystem Play
- Frontop (301313) Deep Dive: From Digital Exhibition Halls to a Physical-AI Data Foundation - FTE Engine, FunSim Simulation and the Gap Between Positioning and Revenue
- Godela Deep Dive: A YC X25 Physics-AI Simulation Startup Whose Ex-Apple, Google and Intel Team Uses Physics-Constrained Surrogate Models to Cut Simulation from Days to Seconds, at Seed Stage with Pilot-Only Access
- HollyWave (688682) Deep Dive: RDSim AI Electromagnetic Simulation and Engineering Agents Close the Simulation-Measurement-Product Loop
- IAT (Alt) Deep Dive: China's First Listed Independent Automotive Design Company Pivots to AI-native R&D with the Yuanfang, Yufeng and Taiyi 3.0 Product System and an NVIDIA Omniverse Partnership, Reporting RMB 975M Revenue and a Widened RMB 250M Net Loss in 2025
- Inductiva Deep Dive: A Closed Case in Open-Solver Orchestration, Eight Premises That Must Be Corrected, and Where Co-Founder and CTO Hugo Penedones Went Next
- Intesim (INTESIM) Deep Dive: A Home-grown Chinese CAE Leader Founded in 2009 with Self-developed Solvers for Structure, Fluid, Electromagnetics, Thermal and Acoustics and Multiphysics Coupling, a ~RMB 100M B Round from CNII and Shenzhen Capital, and 2025 Revenue and Bookings Both above RMB 100M
- iSoftStone (301236) Deep Dive: A ChiNext-Listed IT Services Giant Whose TianShu Platform Builds a Three-Layer Physical-AI Stack with Self-Developed TwinHub Digital Twin, FSim Industrial Simulation, CAE Multiphysics and Phyxis Embodied Control
- Jua Deep Dive: The Earth Physics Transformer as a Large Physics Model for Energy Trading
- Keysight (with ESI Group) Deep Dive: Virtual Prototyping and Hybrid AI Bridge Measurement and Simulation, with Natural-Language Assistants in EDA
- Lightwheel Deep Dive: A Physical AI Data and Evaluation Infrastructure Startup Founded by a Former NVIDIA Simulation Lead, Seven Rounds Totalling Roughly CNY 2.65 Billion at a Valuation Above USD 2 Billion, Seats on Both the Newton and EgoVerse International Standards Committees, With Two Incompatible Accounts of Q1 Orders
- Luminary Cloud: The Physics AI Model Factory Backed by NVentures
- MathWorks Deep Dive: MATLAB/Simulink and Simscape Multiphysics Power Physical AI, with R2026a Copilots and an Agentic Toolkit for Edge Intelligence
- Maniformer (Mifeng) Deep Dive: An Agibot-Spun Physical-AI Data Infrastructure Firm Whose Bodyless MEgo Capture Hardware Is in Mass Production and Which Closed Three Rounds in Half a Year Led by China Telecom
- Mistral AI: How Physics AI Moved From Language Models to EUV Lithography and Nuclear Power
- Monolith AI: The British Engineering-AI Firm That Cut Physical Test Counts Instead of Accelerating Simulation, Now Inside CoreWeave
- Monumo Deep Dive: A Cambridge Deeptech Company Redesigning Electric Motors with the Anser AI Engine, Tens of Millions of Multiphysics Simulations per Day, GBP 10.5M Raised in Total, and a Series A Being Raised
- NAVASTO: Moving Solver Insight Onto the Designer's Desk, and How Autodesk Bought That Entry Point
- Navier: Handing CFD Workflows to Agents, a Seed-Stage Bet on AI for Hardware Engineering
- Neural Concept: An EPFL Spin-out Betting on Neural Surrogates, with $138M Raised
- NP Co. (NP Company) Deep Dive: A 2025-founded Paris Inria-born AI-native simulation startup that raised a EUR 6M pre-seed led by Partech with Mistral AI co-founders as angels, claiming a pre-trained physics foundation model at 1,000x speedup — most claims unverified
- NVIDIA Deep Dive: Building the Full-Stack Physical-AI Infrastructure with PhysicsNeMo, Warp, Cosmos World Foundation Models and Omniverse
- Physical Intelligence Deep Dive: A $2.1B Bet on the GPT Moment for Robotics, With pi0.7 Self-Reporting 60-80 Percent Success on Unseen Tasks and Neither Pricing nor Revenue Disclosed for the Pi API
- Physical Superintelligence Deep Dive: A $58M Seed Bet on Virtual Physicists, With Emmy Targeting AI Data Center Multiphysics Optimization as Its First Commercial Proof Point
- PhysicsX: Pretrained Geometry and Physics for Industrial Engineering at a $2.4B Mark
- Prometheus: A $41B Mark, $18.2B Raised, and Not Yet a Product
- PTC Deep Dive: Creo/Onshape and the ThingWorx Digital Thread Power Physical AI, with NVIDIA Isaac Sim Bridging Robotics Design to Simulation
- Qianjue Technology Deep Dive: How a Body-Agnostic Embodied Brain Pairing the Predictive World Model Polibrain with Brain-Inspired Partitioning and Polynomial Representation Reached a Hundred Thousand Cleaning and Service Robots
- Quanscient Deep Dive: Cloud-Native and Quantum-Powered Multiphysics Simulation as the Data Engine for AI-Native Hardware Engineering
- Rescale Deep Dive: An HPC Operating System and Desktop Engineering Workbench in the Cloud, a $115M Series D, and Five Misread Facts
- Schrödinger Deep Dive: Physics-Based FEP+ Simulation and Machine Learning Power Molecular Discovery, with the 2026 Bunsen Agentic Co-Scientist
- Secondmind Deep Dive: A Cambridge Engineering-AI Firm Whose Bayesian Active Learning Cuts CAE Simulation and Test-Bench Runs by up to 90%, with Roughly USD 73M Raised and Mazda Both Customer and Investor
- Shichuang Technology Deep Dive: SupreCAST Cloud-Native Material-Processing Simulation and Industrial Agents Advance the CAE+AI Strategy
- ShonCloud (Shunyun) Deep Dive: Meshless Particle-Method CFD for EV and Nuclear Fluid Multiphysics
- Shuzhi Maijin Deep Dive: A Shenzhen Fully Self-Developed SNN-plus-LBM CFD Startup Claiming Second-Level Results on Hundreds-of-Millions of Grids on Consumer Hardware, With a Tens-of-Millions-RMB Angel Round From a Ministry-of-Finance-Affiliated Fund and an FPGA Migration Plan
- Siemens (incl. Altair) Deep Dive: The 2025 Altair Acquisition Unites Simcenter and Altair, with PhysicsAI Geometric Deep Learning and PhysicsAI Generate as the Physical-AI Core
- Silurian Deep Dive: A Generative Forecasting Transformer Earth-System Foundation Model for Weather Risk on Energy and Infrastructure Assets
- SimScale Deep Dive: A Cloud CAE Platform Built on Core-Hour Subscriptions, EUR 52M of Equity Kept Separate from EUR 25M of Debt, and Five Misread Facts
- Skild AI Deep Dive: A General Robot Foundation Model Incubated at CMU's Robotics Institute, With a USD 1.4B Series C at a USD 14B Valuation, a Conflicting Series B Size, and an ARR Figure That Is a Run-Rate Rather Than Revenue
- Songying Deep Dive: A Huawei-Cloud-Alumni Startup Building a Domestic Physical AI Operating System, With ORCA OS Positioning Against NVIDIA Omniverse, Three Incompatible Financing Narratives, and a Verifiable Third-Party Anchor in a CNY 20 Million Minority Investment
- Suochen Technology: The Pain of Becoming a Physics-AI Platform from a CAE Vendor
- Synopsys (Ansys) Deep Dive: The 2025 Ansys Acquisition Turns the EDA Leader into a Silicon-to-Systems Engineering Powerhouse, with Ansys SimAI and the NVIDIA Physical-AI Partnership as the Core Physics-AI Levers
- TenFong Deep Dive: An Academician-Led Chinese Industrial Simulation Software Vendor Whose Official Timeline and Database Records Use Incompatible Round Names, With an Official Financing PDF and an Investment Database Both on File
- Tianfu (Nanjing Tianfu Software) Deep Dive: A 2011-founded Chinese Intelligent-CAE Veteran That Closed a Several-Hundred-Million-RMB Series C, Pivoted to an Industrial World Model in 2024, With Unreconcilable Financing and Valuation Records and No Disclosed Revenue
- Tongyuan Soft Control Deep Dive: MWORKS (Syslab/Sysplorer) as a Domestic MATLAB/Simulink Alternative, with MWORKS AI and the MOGICK Engineering Agent Advancing Home-Grown Physical AI
- Vinci AI: Turning Physics Simulation from a Checkpoint into a Continuous Constraint
- Westlake Robotics Deep Dive: Westlake University's First Robotics Spin-out Builds a Dual-Pretrain Brain-and-Cerebellum Architecture with WR1 Whole-Body Foundation Model and GAE Behaviour Expert
- World Labs: Fei-Fei Li's Spatial-Intelligence Unicorn, Acquired by AMD for $8.2B
- XVERSE Deep Dive: A Simulation-World Route to Physical AI Built on the XVERSE Model Family, an Edge-Cloud 3D Engine and the Physics-Based MotionGen Model
- SimWE: From In-House Solvers to a Physical-AI Engine, the Domestic CAE Leader and What Comes Next
- Yunjing Zhifang Deep Dive: A Peking-University-Incubated Cloud-Native GPU CAE Company with the Simversus Platform and the YJ-DORGP Small-Sample Intelligent Solver, Reporting 275 Percent Order Growth, 160 Percent Revenue Growth and Breakeven in 2025
- Zhongke Diwiji (Five Ages) Deep Dive: A CAS-Institute-of-Automation and Tsinghua Spin-out Whose FAM Embodied Foundation Model Uses Action Optical Flow for Cross-Embodiment Generalisation, Reaching 97% Success from 3 to 5 Demos, with EraBrain and Jiling Robots Serving Sinopec, Leapmotor and Midea
- ZWSOFT (688083) Deep Dive: Overdrive Kernel and ZW3D Wukong 2027 Underpin an Integrated CAD/CAE/CAM Stack, with a CAx Agent Platform as the Practical Route to Engineering AI
This report is compiled from public sources. It is for reference only and does not constitute investment advice. Data is current as of the publication date; please refer to the company’s official filings.