AgiBot (Zhiyuan Robotics): Company Profile

Founding

AgiBot — Zhiyuan Robotics in Chinese — was founded in February 2023 in Shanghai by two former Huawei engineers: Deng Taihua, now chairman and CEO, and Peng Zhihui.

Peng is the reason the company was famous before it had a product. He came out of Huawei's “Genius Youth” programme, the scheme that recruits a handful of exceptional graduates a year on salaries reported around two million yuan, and he is separately well known in China as “Zhihui Jun”, an engineering personality whose home-built robotics projects — a working mechanical prosthetic arm among them — drew a large online following. He left Huawei in December 2022; the company existed two months later.

Three years old, and it now accounts for roughly two in every five humanoid robots shipped anywhere. Nothing else in this section has moved at that speed.

Country and Headquarters

AgiBot is a Chinese company headquartered in Shanghai, operating as Zhiyuan Innovation (Shanghai) Technology Co., Ltd.

Leadership

Deng Taihua is chairman and CEO, having previously held senior positions at Huawei. Peng Zhihui is co-founder and the company's technical figurehead. The pairing is deliberate and effective: an executive who has run large organisations, and an engineer with a public following that the company has used as a recruiting and marketing asset in a way no Western humanoid maker has matched.

Company Size and Funding

AgiBot has raised across more than ten rounds in three years, from an investor list that reads like a map of Chinese technology and manufacturing: Tencent, BYD, HongShan (the former Sequoia China), Hillhouse, JD.com, SAIC Motor, the Shanghai Embodied Intelligence Fund, and more recently LG Electronics and Mirae Asset from South Korea. Its valuation passed 15 billion yuan in a Tencent-led round in March 2026 and has since been reported above 20 billion.

Two routes to a stock market at once

The capital-markets manoeuvring is the most distinctive thing about this company after its shipment numbers, and it is barely covered outside China.

In 2025 AgiBot moved to take control of Swancor Advanced Materials, a Shanghai-listed maker of wind-turbine blade materials, buying 63.62% through two affiliated partnerships for a sum reported between $279 and $290 million. The change of control completed in September 2025 and the stake purchase in November. This was the first time an embodied-AI company took control of an A-share listed company, and AgiBot's chairman became Swancor's de facto controller. The company publicly denied it was a back-door listing, arguing the transaction does not meet the regulatory definition.

Then, in July 2026, it began a separate Hong Kong IPO process with CITIC Securities as sponsor, reportedly targeting a valuation around US$20 billion and telling the market it expects roughly 4 billion yuan of revenue for the year.

Whatever the Swancor deal is formally called, the combined effect is a company three years old with a controlling position in a mainland-listed vehicle and a Hong Kong listing in progress. That is a different relationship with capital markets than any Western humanoid maker has, and it is worth understanding before comparing valuations across the two.

Development Directions and Products

AgiBot's catalogue is wider than anyone else's here, and deliberately not all bipedal:

• Yuanzheng A2 — the flagship full-size humanoid, around 169–175 cm, modular with customisable exteriors, aimed at reception, exhibition, customer service and light industrial work. Variants include A2-Max for heavier tasks and A2-W on a wheeled base.
• Lingxi X1 and X1-W — a 130 cm, 33 kg research platform with 34 degrees of freedom, open-sourced globally in hardware documentation and software. A company shipping more humanoids than anyone else also publishes one for others to copy.
• Genie G1 / G2, the C5 cleaning robot and the OmniHand end-effectors round out the range.

A figure worth keeping straight: of roughly 15,000 embodied robots produced by June 2026, about 7,000 are bipedal. The rest are wheeled bases and other form factors. Headline totals across companies are often not comparing the same thing.

Genie Operator, and AgiBot World

Unlike Apptronik, which runs Google's models, AgiBot builds its own brain. Genie Operator-1 combined a vision-language-action model with a mixture-of-experts architecture; GO-2 followed on 9 April 2026.

The more consequential release was free. In December AgiBot open-sourced AgiBot World, a manipulation dataset of over one million real robot trajectories across 217 tasks in five deployment settings, collected by more than a hundred identical robots with human verification, published on HuggingFace and GitHub. Training data is the genuine bottleneck in this field — Figure expects to spend around a billion dollars a year buying it — and AgiBot gave a million trajectories away. The associated paper was a best-paper finalist at IROS 2025.

Markets and Customers

AgiBot's claim rests on volume rather than on marquee logos. It passed its 1,000th robot in January 2025, shipped more than 5,100 humanoids during 2025, reached its 15,000th embodied robot by June 2026, and reports 8,400 humanoid shipments in the first half of 2026 alone — about 44% of the global total. No other company in this section is within an order of magnitude on units.

The honest qualification is what those robots do. A large share go to universities, laboratories, developers, exhibitions and retail showrooms rather than onto production lines, so this is not directly comparable with Agility's 65,000 logged operating hours or UBTECH's audited factory revenue. Shipping the most robots and having them do the most work are different achievements, and AgiBot currently has the first.

What the volume does buy is real: a manufacturing learning curve, component pricing, and — through AgiBot World — a data flywheel that a company shipping dozens of robots cannot run. Those advantages compound quietly, and they are the reason this three-year-old company is worth watching more closely than its customer list suggests.