AgiBot
Last reviewed
May 9, 2026
Sources
No citations yet
Review status
Needs citations
Revision
v8 · 6,650 words
Improve this article
Add missing citations, update stale details, or suggest a clearer explanation.
Last reviewed
May 9, 2026
Sources
No citations yet
Review status
Needs citations
Revision
v8 · 6,650 words
Add missing citations, update stale details, or suggest a clearer explanation.
| AgiBot | |
|---|---|
| General information | |
| Full name | AGIBOT Innovation (Shanghai) Technology Co., Ltd. |
| Chinese name | 智元机器人 (Zhiyuan Jiqiren) |
| Founded | February 2023 |
| Founders | Peng Zhihui (CTO), Deng Taihua (CEO), Yan Weixing |
| Headquarters | Shanghai, China |
| Industry | Robotics, Embodied AI |
| Products | Humanoid robots, quadruped robots, wheeled service robots, dexterous hands |
| Revenue (2025) | 1.05 billion yuan (US$147 million) |
| 2025 shipments | 5,168 humanoid robots (No. 1 globally per Omdia) |
| Cumulative output (Mar 2026) | 10,000+ robots |
| Valuation (target) | HK$40 billion to HK$50 billion (US$5.1 billion to US$6.4 billion), 2026 IPO target |
| Website | agibot.com |
AgiBot (formerly known as Zhiyuan Robotics; Chinese: 智元机器人, Zhiyuan Jiqiren) is a Chinese robotics company headquartered in Shanghai that designs, manufactures, and deploys humanoid robots, wheeled service robots, quadruped robots, and dexterous robotic hands for industrial, commercial, research, and consumer applications. The company was founded in February 2023 by Peng Zhihui, a former Huawei "Genius Youth" engineer better known on Bilibili as Zhihui Jun (稚晖君), together with longtime Huawei executive Deng Taihua. AgiBot reached a cumulative production milestone of 10,000 robots on 30 March 2026, becoming one of the first companies in the industry to reach that scale. Its robots are deployed across automotive, electronics, logistics, retail, hospitality, and education sectors in China and internationally.[1][2][3]
AgiBot's technical approach combines proprietary hardware (high-torque actuators, the OmniHand dexterous hand family, and multimodal sensor suites) with a vertically integrated artificial intelligence software stack that includes the GO-1 and GO-2 generalist embodied foundation models, the Lingqu OS embodied operating system, the Genie Envisioner world-model platform, and the open-source AgiBot World manipulation dataset. The company collaborates closely with OpenDriveLab (affiliated with the University of Hong Kong and the Shanghai AI Lab) on embodied intelligence research and has open-sourced major components of its software stack on GitHub and Hugging Face.[4][5]
AgiBot ranked first globally in humanoid robot shipments in 2025 with 5,168 units delivered, capturing approximately 39 percent of the global market according to research firm Omdia. The company is preparing a Hong Kong IPO targeted for the third quarter of 2026 at a valuation of HK$40 billion to HK$50 billion (US$5.1 billion to US$6.4 billion).[2][6]
AgiBot was founded on 27 February 2023 in Shanghai. Co-founder Peng Zhihui, born in 1993 in Ji'an, Jiangxi Province, graduated from the University of Electronic Science and Technology of China (UESTC) in Chengdu with a master's degree in 2018. While still a student he developed a bipedal robot capable of obstacle avoidance and autonomous navigation, winning an innovation award at the 2018 World Robot Olympiad. He briefly worked at smartphone maker Oppo as an algorithm engineer before joining Huawei in 2020 through the company's selective "Genius Youth" (天才少年) recruitment programme launched by founder Ren Zhengfei in 2019. Peng was placed on the highest of three salary tiers at roughly 2 million yuan per year and worked as an AI algorithm engineer in the computing product line, contributing to projects in AI edge heterogeneous computing and the Ascend AI platform. Outside Huawei he became one of the most popular technology creators on Bilibili under the handle Zhihui Jun (稚晖君), with more than 2.5 million followers and viral videos showcasing self-balancing robots, an Iron Man-style robotic arm, a self-riding bicycle, and a robotic arm capable of stitching grape skins. Peng announced his departure from Huawei on Bilibili in late December 2022.[7][8][9]
Co-founder Deng Taihua brought 27 years of Huawei experience to the venture. He joined Huawei in 1995 and rose to become Vice President of the company and President of the Computing Product Line, where he led ecosystem development for the Kunpeng processor and Ascend AI chip families and grew the Ascend developer community to over two million members by 2020. Deng left Huawei at the end of 2022 and was formally appointed Chairman and CEO of AgiBot in March 2025, replacing the company's earlier interim leadership. A third early co-founder, Yan Weixing, also from a hardware engineering background, joined the founding team but later took a lower public profile.[10][11]
In August 2023, six months after its founding, AgiBot unveiled its first humanoid robot, the RAISE A1 (also marketed as Yuanzheng A1, 远征 A1). The 175 cm, 53 kg bipedal platform featured 49 degrees of freedom, an in-house perception stack combining LiDAR and RGB-D cameras, a proprietary task-execution system called "WorkGPT", and an "Embodied Intelligent Brain" (EI-Brain) architecture for autonomous task planning. The launch event in Shanghai showcased the robot tightening bolts, performing vehicle inspections, and conducting laboratory experiments, positioning AgiBot from the outset as an industrial-application company rather than a consumer-novelty maker.[12]
AgiBot's earliest backers included three of China's most prominent technology investors. The seed round, completed in September 2023, drew capital from HongShan (the standalone successor of Sequoia China after its July 2023 split from Sequoia Capital), Hillhouse Investment (also called Hillhouse Capital), and BlueRun Ventures (BRV Partners). Carmaker BYD and JD.com's strategic arm joined a strategic round shortly afterwards, marking some of the earliest equity tickets that automotive original equipment manufacturers wrote into the Chinese humanoid sector. Cumulative seed-stage proceeds reportedly reached US$83.8 million.[13][14]
In January 2024, AgiBot established its dedicated humanoid manufacturing facility in Shanghai's Lingang Special Area. In August 2024 the company unveiled five new commercial robots at a launch event branded "Hello, A2": the bipedal Yuanzheng A2 (the production successor to the RAISE A1), the heavy-duty A2-Max, the wheeled dual-arm A2-W, and two compact Lingxi-series humanoids, the X1 and X1-W.
In September 2024 AgiBot opened the AIDEA Giga Data Factory, a purpose-built facility spanning over 4,000 square meters that simulates five core scenarios (home, dining, industry, supermarket, and office) using more than 3,000 real objects. Inside, almost 100 tele-operated humanoids generate 30,000 to 50,000 robot-action data points each day for training the company's foundation models. By 15 December 2024, AgiBot had manufactured 962 units and publicly announced the start of mass production, opening its production line to media and disclosing per-line throughput at the Shanghai factory.[15][16]
In January 2025 AgiBot rolled out its 1,000th general-purpose embodied robot. Throughout the year the company stacked technical, commercial, and political milestones at an unusually rapid pace:
AgiBot shipped 5,168 units in 2025 according to Omdia's General-Purpose Embodied Intelligent Robot 2026 report (published January 2026), capturing roughly 39 percent of the global humanoid robot market and ranking first worldwide in shipments for the year. Total annual industry shipments came to about 13,000 humanoid units. Revenue jumped from 60 million yuan in 2024 to 1.05 billion yuan (about US$147 million) in 2025, a roughly 17-fold expansion.[2][29]
In January 2026 AgiBot made its U.S. market debut at CES 2026 in Las Vegas, displaying the A2 Series, X2 Series, G2 Series, and D1 quadruped series alongside the Genie Sim 3.0 simulation platform. The company won "Best of CES 2026" awards from Ubergizmo and MacStories and a Netzwelt Innovation Award 2026, and inaugurated its first overseas robot experience centre in Malaysia.[6][30]
In February 2026, AgiBot ran AGIBOT NIGHT, a 60-minute live gala in Shanghai in which more than 200 humanoid robots performed synchronised dance, magic, comedy, music, and martial-arts routines under broadcast conditions. AgiBot described it as the world's first large-scale, full-length live event led by humanoid robots. Days later the company unveiled the next-generation Expedition A3 humanoid and the GO-2 (Genie Operator-2) foundation model.[31][32]
By 30 March 2026, AgiBot announced the rollout of its 10,000th robot. The acceleration was dramatic: nearly two years to reach 1,000 units, roughly one year to scale from 1,000 to 5,000, and just three months to scale from 5,000 to 10,000. On 15 April 2026, AgiBot and Longcheer announced what they described as the world's first deployment of embodied AI in a consumer-electronics precision-manufacturing mass-production line, with multiple G2 robots running tablet testing on a live Longcheer assembly floor; AgiBot said it expected to expand the deployment to about 100 robots by the third quarter of 2026.[1][33]
AgiBot completed its corporate restructuring from a limited-liability company to a joint-stock company in late 2025, a procedural step typically preceding a public listing. The company appointed China International Capital Corporation (CICC) and CITIC Securities as joint sponsors for its planned Hong Kong IPO earlier in 2025, with Morgan Stanley joining the syndicate later that year. The listing is targeted for the third quarter of 2026 at a valuation of HK$40 billion to HK$50 billion (about US$5.1 billion to US$6.4 billion); AgiBot plans to sell 15 to 25 percent of its shares, potentially raising more than US$1 billion.[34]
| Name | Chinese | Role | Background |
|---|---|---|---|
| Deng Taihua | 邓泰华 | Chairman and CEO (from March 2025) | 27-year Huawei veteran; former Vice President of Huawei and President of the Computing Product Line; led Kunpeng and Ascend ecosystems; left Huawei end of 2022.[10] |
| Peng Zhihui | 彭志辉 (Zhihui Jun, 稚晖君) | Co-founder, Chief Innovation and Technology Officer | Born 1993, Ji'an, Jiangxi; UESTC master's 2018; former Oppo algorithm engineer; Huawei "Genius Youth" 2020 to 2022 in the computing product line; influential Bilibili maker with 2.5 million-plus followers.[7][8] |
| Yan Weixing | 闫维新 | Co-founder, hardware engineering | Mechanical and actuator-systems engineer who joined the founding team in early 2023.[11] |
AgiBot produces a broad portfolio spanning full-size bipedal humanoids, compact humanoids, wheeled service platforms, miniature personal humanoids, quadruped robots, and dexterous robotic hands. The product lines fall under three Chinese brand families: Yuanzheng (远征, "Expedition", the A-series), Lingxi (灵犀, "Spirit", the X-series), and Jingling (精灵, "Genie", the G-series). The D-series quadrupeds and Q-series miniature humanoid round out the lineup.
| Model | Type | Height | Weight | DOF | Max speed | Payload | Key use cases | Status |
|---|---|---|---|---|---|---|---|---|
| RAISE A1 (Yuanzheng A1) | Bipedal humanoid | 175 cm | 53 kg | 49 | 7 km/h | 80 kg | Industrial automation, inspections | Prototype (2023) |
| Yuanzheng A2 | Bipedal humanoid | 175 cm | 55 kg | 49 | 11.9 km/h | 15 kg | Service, industrial, research | Production |
| A2 Lite | Bipedal humanoid | 175 cm | ~55 kg | ~40 | 11.9 km/h | 15 kg | Entertainment, commercial | Production |
| A2 Ultra | Bipedal humanoid | 169 cm | ~69 kg | 40 | 11.9 km/h | 15 kg | Service with mapping and autonomy | Production |
| A2 Max | Bipedal humanoid | 175 cm | 85 kg | 67 | 3.6 km/h | 40 kg | Heavy-duty handling, palletising | Production |
| A2-W | Wheeled dual-arm humanoid | 163 cm | 230 kg | n/a | n/a | n/a | Industrial manipulation, logistics | Production |
| Expedition A3 | Bipedal humanoid | 173 cm | 55 kg | 40+ | n/a | n/a | Interactive service, performance, research | Production (Feb 2026) |
| Lingxi X1 | Compact humanoid (open source) | 130 cm | 33 kg | 34 | 3.6 km/h | 0.5 kg | Education, research | Available |
| Lingxi X2 | Compact humanoid | 130 cm | 36 kg | 29 | 5.4 km/h | 1 kg | Research, education, light industrial | Available |
| X2 Ultra | Compact humanoid | 131 cm | 39 kg | 30 | 5.4 km/h | 1 kg | Research, system integration | Available |
| G2 Genie | Wheeled humanoid (industrial) | 175 cm | ~185 kg | 26 | 7 km/h | 5 kg per arm | Industrial automation, 24/7 operation | Production |
| QUESTER1 (Q1) | Mini personal humanoid | 80 cm | 4 kg | n/a | n/a | n/a | Education, research, hobby, personal | Announced |
| D1 Pro / Ultra / Max Pro | Quadruped | varies | varies | n/a | varies | varies | Inspection, security, field operations | Available |
| OmniHand Pro | Dexterous hand | n/a | 750 to 820 g | 19 | n/a | varies | Manipulation research, industrial assembly | Available |
The A2 is AgiBot's flagship full-size humanoid platform and the volume product behind the company's 2025 leadership in shipments. Standing 175 cm tall and weighing 55 kg, the A2 mirrors human proportions and features 49 degrees of freedom. The dual 7-DOF arms deliver industrial-grade repeatability for assembly and material-handling tasks. The robot operates for approximately two hours per charge, walks at up to 3.3 m/s (11.9 km/h), and can handle 15 kg payloads. A 200 TOPS computing system powers real-time AI inference, while the sensor suite includes microphone arrays, LiDAR, and RGB-D cameras. The A2 was the first full-size humanoid to receive simultaneous safety certifications from Chinese, U.S., and European regulators in May 2025, and it carries the company's Guinness World Record for the longest unpowered-off journey by a humanoid robot.[35][20][27]
The A2 family includes several variants tailored to different use cases:
The Expedition A3, unveiled in February 2026, is AgiBot's next-generation full-size bipedal humanoid. Standing 173 cm and weighing only 55 kg, it uses lightweight magnesium, titanium, and TPU materials and is designed around what the company calls a "golden ratio" silhouette. Distinguishing technical features include 10-hour endurance, a 10-second hot-swap battery design, ultra-wideband (UWB) centimetre-level swarm positioning that supports synchronised choreography for up to 100 robots, shoulder tactile sensing, and a 360-degree multi-array microphone for natural multi-speaker interaction. Launch demonstrations included aerial flying kicks, consecutive mid-air strikes, and cyclone spin manoeuvres, showcasing the platform's whole-body dynamic control.[32]
The X1 and X2 are compact, open-source humanoid platforms designed for research and education.
The X1 stands 130 cm tall, weighs 33 kg, and features 34 DOF with proprietary PowerFlow R-series actuators on a CAN-bus communication backbone. It runs on the AimRT middleware (open source, ROS 2 compatible) with reinforcement learning modules for locomotion on Ubuntu 22.04. AgiBot publicly released most design materials, schematics, software-framework source code, middleware, and basic operational algorithms for the X1 free of charge, positioning it as a developer-community platform. The X1 is priced from approximately US$20,000.[39]
The X2 improves on the X1 with refined manipulation and more robust locomotion. At 130 cm and 36 kg with 29 DOF (6 DOF per arm), it walks at 1.5 m/s and offers two hours of battery life. The X2 is offered in base, Pro, and Ultra trims. The X2 Ultra increases DOF to 30, adds 3D LiDAR, stereo RGB cameras, and an RGB-D camera, and can be configured with an NVIDIA Orin NX module providing 157 TOPS of AI compute. List pricing ranges from approximately US$27,300 for the base X2 to over US$43,000 for the X2 Ultra.[40][41]
The G2 Genie, unveiled in October 2025, is an industrial-grade wheeled humanoid built for 24/7 factory and warehouse operation. It stands 175 cm tall with an IP42-rated chassis and IP50-rated arms. The robot features the industry's first cross-shaped wrist 7-DOF force-controlled arm with full joint torque sensors, achieving force-control accuracy better than 0.5 N. A 3-DOF waist and 2-DOF lower body enable bending, twisting, and squatting. The G2 uses dual hot-swappable batteries and supports compute configurations including the Rhino R1 (500 TOPS) and the NVIDIA Jetson Thor T5000 (2,070 TFLOPS), giving it one of the most powerful onboard AI platforms in the industry. It is powered by AgiBot's GO-1 foundation model and GE-1 world model.[26]
In October 2025 AgiBot announced a framework order from electronics ODM Shanghai Longcheer Technology valued in the hundreds of millions of yuan, covering close to 1,000 G2 units. In April 2026, multiple G2 robots were integrated into Longcheer's tablet production lines, deployed primarily at MMIT (Multimedia Integrated Testing) stations to perform precision loading and unloading: autonomously picking up tablets, navigating crowded factory aisles, placing devices into testing fixtures with millimetre accuracy, and sorting finished or defective units. AgiBot called it the world's first deployment of embodied AI in a live consumer-electronics precision-manufacturing mass-production line and said it planned to expand the integration to about 100 robots by the third quarter of 2026.[33]
The QUESTER1, unveiled in late 2025, is a miniature humanoid standing approximately 79 to 80 cm tall and weighing only 4 kg, roughly one-eighth the weight of AgiBot's full-size humanoids. The Q1 features miniaturised Quasi-Direct Drive (QDD) joints smaller than an egg that deliver crash-resistant full-body force control. The robot supports an open-source ecosystem with SDK and HDK access; users can 3D-print custom shells from open files and program motions through a zero-code block-based platform. Out-of-the-box features include voice interaction, English tutoring, dance coaching, and gesture-based tasks, all running on the Agi-Soul AI platform. The Q1 is targeted at education, research, and personal-companion use cases.[42]
AgiBot also produces a line of quadruped robots for outdoor and industrial field applications. The D1 series includes the D1 Pro (compact, dynamic locomotion), the D1 Ultra (industrial-grade for security inspection, emergency response, and research), and the D1 Max Pro (high-capacity platform for industrial inspection, equipment transport, and challenging terrain). The D1 series received a Netzwelt Innovation Award 2026 for its handling of difficult terrain.[43][6]
The OmniHand Pro is AgiBot's flagship dexterous robotic hand, with 19 active degrees of freedom in a compact 750 to 820 g package and human-like proportions. It integrates more than 150 tactile sensors with 0.01 N resolution. A newer flagship variant featuring a 22+3 DOF tendon-driven design weighs only 500 g and offers a 10:1 load-to-weight ratio, full-hand 3D tactile sensing, an integrated palm camera, sub-0.3-second response time, and a wide wrist range of 55 degrees pitch and 40 degrees yaw. OmniHand Pro is sold both bundled into AgiBot's robots and standalone for research and industrial-assembly customers.[44]
AgiBot organises its software and AI portfolio around an architecture it calls "One Robotic Body, Three Intelligences", spanning Locomotion Intelligence (whole-body motion control), Manipulation Intelligence (dexterous task execution), and Interactive Intelligence (multimodal human-robot interaction). The eight foundational AI products under this umbrella are tied together through a unified Physical AI platform driven by data, simulation, and real-world deployment.[44]
Genie Operator-1 (GO-1), launched on 11 March 2025, is AgiBot's first generalist embodied foundation model. It introduces the Vision-Language-Latent-Action (ViLLA) framework, which combines a Vision-Language Model (VLM) with a Mixture of Experts (MoE) architecture. The system has two key components: the Latent Planner, which learns general action understanding from cross-embodiment and human-operation data, and the Action Expert, which uses over one million real robot demonstrations to achieve high-frequency dexterous manipulation. GO-1 is pretrained on the AgiBot World dataset and is published with model weights on Hugging Face and code on GitHub. The company reported that GO-1 achieved more than a 60 percent success rate on complex real-world dexterous and long-horizon tasks, outperforming the prior Robotics Diffusion Transformer (RDT) approach by 32 percentage points.[17]
GO-2 (Genie Operator-2), unveiled in February 2026, succeeds GO-1 with a unified architecture that integrates logical reasoning and action execution within a single network. AgiBot reports the model ranks first on the LIBERO benchmark across Spatial, Object, Goal, and Long-Horizon tasks with an average success rate of 98.5 percent, achieves 86.6 percent zero-shot success on LIBERO-Plus environments with disturbances, scores an average of 47.4 on VLABench, and reaches 82.9 percent real-world success on Genie Sim 3.0 using only simulation-trained data. GO-2 is part of the broader Physical AI platform announced alongside the Expedition A3.[45]
AgiBot World is an open-source large-scale robot manipulation dataset created in collaboration with OpenDriveLab (the embodied-AI and autonomous-driving research initiative associated with the University of Hong Kong and the Shanghai AI Lab). Collected by more than 100 homogeneous AgiBot G1 robots in the 4,000-square-meter AIDEA Giga Data Factory, the dataset contains 1,001,552 trajectories totaling 2,976.4 hours and covering 217 specific tasks, 87 skills, and 106 scenes across five core scenarios: home, dining, industry, supermarket, and office.
The dataset is released in two tiers: AgiBot World Beta (1,003,672 trajectories, approximately 43.8 TB) and AgiBot World Alpha (a curated 92,214-trajectory subset, approximately 8.5 TB). Compared with Google's Open X-Embodiment dataset, AgiBot World features data at roughly ten times the scale, scene coverage expanded by approximately 100 times, and data quality elevated from laboratory-level to industrial-level standards. The dataset has been described as the "ImageNet moment" for embodied AI. The corresponding paper, AgiBot World Colosseo: A Large-scale Manipulation Platform for Scalable and Intelligent Embodied Systems (arXiv 2503.06669), was named a Best Paper Award finalist at IROS 2025 and accepted for publication in IEEE Transactions on Robotics (T-RO) in 2026. The dataset is available for non-commercial research use on GitHub and Hugging Face.[5][46]
Lingqu OS, released at WAIC on 27 July 2025, is AgiBot's embodied intelligent operating system. It comprises three layers:
Lingqu OS rolled out under an open co-development model in the fourth quarter of 2025, with AgiBot positioning the system as a horizontal platform that other robotics manufacturers and integrators can adopt.[23]
Genie Envisioner (GE), introduced in August 2025, is a unified world-model platform that combines robot sensing, policy learning, and simulation evaluation in a single closed-loop video-generative architecture. The full system has three components: GE-Base (the World Foundation Model that predicts future video frames), GE-Act (the World Action Model that produces 54-step torque trajectories within 200 ms on a commodity GPU), and GE-Sim (the World Simulator), accompanied by the EWMBench evaluation framework. AgiBot reported that, on the AgiBot G1 platform, GE-Act consistently outperformed state-of-the-art VLA baselines including UniVLA and NVIDIA's GR00T N1 on complex manipulation tasks such as making a sandwich and pouring tea. The Genie Envisioner technical report (arXiv 2508.05635) was released on 8 August 2025.
Genie Envisioner 2.0, announced in late 2025, evolves GE from a video predictor into an interactive world simulator (GE 2-Sim) that supports closed-loop interaction with simulated environments. Code and assets are open-sourced via the AgibotTech organisation on GitHub.[47][25]
AgiBot's broader software ecosystem includes:
AgiBot has completed at least eight disclosed funding rounds in its first three years, attracting many of China's most prominent technology investors plus international strategic capital from LG Electronics and Mirae Asset.
| Round | Date | Amount | Lead and key investors |
|---|---|---|---|
| Seed | September 2023 | ~US$83.8 million (cumulative seed) | HongShan, Hillhouse Investment, BlueRun Ventures, BYD (strategic) |
| Series A | 2024 | Undisclosed | CDH Investments, iSoftStone, Sanhua Holding Group, SAIC Motor, JD.com |
| Series B | 24 March 2025 | Hundreds of millions of yuan | Tencent (lead), BlueRun Ventures, Lanchi Ventures, Longcheer Technology, Wolong Electric Drive, Huafa Group, Huajin Capital |
| Series B extension | 1 August 2025 | Undisclosed | LG Electronics, Mirae Asset (Mirae Asset-LG Electronics New Growth Fund) |
As of September 2024, the company's valuation stood at approximately US$982 million according to PitchBook. Following the Tencent-led Series B in March 2025, valuation reached approximately 15 billion yuan (about US$2.07 billion). For its planned Hong Kong IPO, AgiBot is targeting a valuation of HK$40 billion to HK$50 billion (US$5.1 billion to US$6.4 billion), with CICC, CITIC Securities, and Morgan Stanley as joint sponsors. The company plans to sell 15 to 25 percent of its shares, potentially raising more than US$1 billion. The preliminary prospectus is expected to be filed in early 2026, with a public listing targeted for the third quarter.[34][14]
In July 2025, AgiBot moved separately to acquire a controlling stake (at least 63.62 percent) in Swancor Advanced Materials, a Shanghai-listed manufacturer of corrosion-resistant materials and wind-turbine blade components, for roughly 2.1 billion yuan (about US$294 million). Swancor's stock subsequently hit eleven consecutive daily limit-up moves, breaking A-share records and becoming the first 10-bagger of the year, fuelling speculation about a back-door listing. AgiBot publicly denied that the deal represented a reverse-IPO attempt and noted it could not satisfy China's three-year operating-history requirement for such transactions.[21][22]
AgiBot operates a humanoid manufacturing facility in Shanghai's Lingang Special Area that entered mass production in late 2024. Production capacity has scaled rapidly:
| Date | Cumulative units produced |
|---|---|
| 15 December 2024 | 962 |
| January 2025 | 1,000 |
| December 2025 | 5,000 |
| 30 March 2026 | 10,000 |
It took AgiBot nearly two years to reach the first 1,000 units, roughly one year more to scale to 5,000, and only about three months to double again to 10,000, a more than fourfold acceleration in production speed.[1][28]
AgiBot's revenue grew from 300,000 yuan (about US$42,000) in its founding year of 2023, to 60 million yuan (about US$8.4 million) in 2024, to 1.05 billion yuan (about US$147 million) in 2025, a roughly seventeen-fold expansion year-over-year. AgiBot has disclosed a publicly stated objective of reaching about 10 billion yuan (US$1.4 billion) in revenue by 2027 as part of an internal three-stage "358" plan tied to the company's commercialisation roadmap.[29]
AgiBot operates in an increasingly crowded Chinese humanoid robotics market and faces a small but ambitious set of Western rivals.
According to a TrendForce report from 9 April 2026, China's humanoid robot output is projected to surge 94 percent in 2026, with Unitree and AgiBot expected to capture nearly 80 percent of the global market combined. Both companies are scaling: Unitree has committed to capacity of 75,000 humanoid robots and 115,000 quadrupeds per year; AgiBot is preparing for the IPO funding round to underwrite the next phase of factory expansion.[51]
| Company | 2025 humanoid shipments | Lead investor(s) | Flagship robot | HQ |
|---|---|---|---|---|
| AgiBot | 5,168 | Tencent, HongShan, BYD, LG Electronics | A2 / Expedition A3 | Shanghai |
| Unitree | ~5,500 (per Unitree); ~4,200 (per others) | Geely-linked funds, Tencent (later) | G1 / H1 / R1 | Hangzhou |
| UBTech | ~1,000 | Public (Hong Kong) | Walker S2 | Shenzhen |
| Tesla Optimus | low thousands (estimated) | Tesla (in-house) | Optimus Gen 3 | California |
| Figure AI | low hundreds (estimated) | Microsoft, NVIDIA, OpenAI Startup Fund | Figure 03 | California |