Editor’s note: This article is republished from Jumpstart, a TechNode Group publication.
The robot could follow the route. The worker knew when to skip it.
Around 2022, Microintelligence encountered a problem that exposed the difference. A customer needed to inspect and grind a large die-cast component more than two metres long. The part required close to 3,000 imaging positions, and the customer wanted the process completed within ten minutes.
According to chairman and chief executive Zhang Zhiqi, a conventional robot following a fixed grinding path could take an hour. A skilled worker needed only seven or eight minutes.
The problem was not simply speed. A worker looked for imperfections and concentrated on the places that needed attention. The programmed machine followed its route.
For Microintelligence, the assignment pushed an inspection business toward a harder ambition: making what a machine sees determine what it does. Zhang described the project in a published interview; the customer was not identified, and the timing figures are his account rather than independently verified benchmarks. Interview with Zhang Zhiqi
That distinction—between finding a problem and acting on it—runs through the company’s development. Founded in 2018 in Changzhou, a manufacturing city in eastern China, Microintelligence began with AI-powered quality inspection. It has since combined sensing, software and robotic control in products that inspect, handle and finish manufactured parts.
Its proposed Hong Kong listing now brings that transformation into view. The company published its post-hearing information pack on October 9, but its more revealing milestones happened inside the business: learning to inspect difficult surfaces, acquiring another technology company, and making robots a larger part of what it sells. Hong Kong listing document
First, Teach the Machine to Look
Microintelligence’s early opportunity was less expansive than its current ambition. It helped manufacturers check whether their products were good enough to ship.
The task was repetitive, but not necessarily simple. A defect might be small, its appearance might vary, and the surface carrying it might be difficult to photograph. A dependable system needed the right lighting and image capture as well as an algorithm capable of interpreting the result.
An early case study published by technology partner Baidu described Microintelligence’s surface-inspection equipment for small manufactured products. The equipment used machine vision and deep learning to locate and classify defects, with cameras covering multiple sides of an object. Baidu identified lighting and algorithms as two central difficulties in building the system. Baidu’s Microintelligence case study
This was a useful place for an AI company to begin. The customer’s question was concrete: could the equipment perform an inspection task reliably enough to justify buying it?
It also gave Microintelligence an education that a software demonstration could not. Images had to be captured under real production conditions. Outputs had to fit a factory’s quality requirements. Hardware and software had to work together.
The company was developing not just the ability to recognize flaws, but the equipment needed to make that ability useful.
The Part That Changed the Question
The large casting project demanded another step.
Die casting produces metal components by forcing molten metal into a mould. Once formed, a part may still need inspection and finishing. Excess material along an edge—a burr, for example—can require removal. Crucially, the location and extent of that work can vary from part to part.
For Microintelligence, identifying the imperfection was therefore only half the assignment. The system needed to translate the inspection into a plan for working on that particular component.
The approach brought together capabilities the company had been developing: capturing images while a robotic arm remained in motion, locating defects, and generating positions and movement paths through software. Rather than treating inspection and grinding as disconnected operations, it sought to connect the information from one with the action of the other.
A 2024 report from the China Electronics Standardization Institute documented this link in Microintelligence’s systems: inspection software recorded and analysed defects, then generated positions and paths for a grinding robot to repair the affected products. Machine-vision industry report
The significance was not that every detected defect could be repaired. It was that the machine’s next move could depend on what the inspection found.
That gave Microintelligence a more demanding role in the factory. It was no longer enough to deliver an answer about a part. The company also had to help deliver an operation on it.
A Different Company Takes Shape
As that technical direction developed, the organization changed too.
In December 2022, Microintelligence acquired Jiangsu Zhiyun Tiangong. Its listing document describes the transaction as a turning point: a company previously driven mainly by AI-enabled equipment expanded strategically into embodied-intelligence industrial robots.
The acquisition also joined two leadership paths. Pan Zhengyi had been with Microintelligence since its establishment and became chief operating officer in December 2022. Zhang, who founded Zhiyun Tiangong in 2021, became Microintelligence’s chairman in January 2023 and now serves as chairman and chief executive. His previous career included roles at SAP and Baidu’s intelligent cloud business.
The filing does not attribute every subsequent product advance to the acquisition. It does, however, make clear that the transaction helped reshape the business and its product mix. Company history and management disclosures
Microintelligence’s growth was therefore not simply a matter of its founding team gradually adding features. The company assembled a broader organization around the move from inspection equipment toward robot systems.
More Than a Pair of Eyes
Today, that direction is visible in Microintelligence’s TRON product range.
The company offers industrial inspection robots, composite robots and systems combining inspection with grinding. These sit alongside AI-enabled equipment and specialized sensing modules, with applications across electronics, automotive manufacturing, energy-related industries and semiconductors. Official product portfolio
The grinding system provides a particularly clear connection to the company’s origins. Inspection remains essential, but its output becomes an instruction for further work.
This is what makes Microintelligence’s use of “embodied intelligence” more understandable than the term initially suggests. Its industrial machines do not need to resemble people. They need to connect perception with a useful physical operation.
Nor does that make them universally capable. A system designed to inspect and finish a casting is not automatically ready for every task elsewhere in the factory. The company’s opportunity lies in extending capabilities across related jobs while meeting the specific requirements of each.
The product portfolio reflects that progression: the original inspection problem has become the foundation for a wider set of machines.
The Business Follows the Machine
The transformation also appears in sales.
Microintelligence’s embodied-intelligence industrial robot products accounted for 26.3% of revenue in 2023. By 2025, they contributed 57%, becoming the company’s largest revenue category. Total revenue increased from RMB434 million to RMB795.5 million over that period.
The financial picture is less uniform than the growth curve. After a net loss in 2023, the company reported profits of RMB15.7 million in 2024 and RMB5.1 million in 2025. Its expansion has become commercially substantial, but profitability has not risen steadily alongside revenue. Financial disclosures
Those numbers establish how much the business has changed without settling what comes next. Microintelligence must still turn the capabilities developed through industrial projects into products whose commercial value survives the cost of building and delivering them.
Its Hong Kong listing process is a milestone in that effort, not its conclusion.
The casting assignment captures the company’s journey better than an expansive promise about the future of AI. A machine could already move along a prescribed route. Microintelligence’s task was to make the route respond to the part in front of it.
It began by helping factories find the flaw. Increasingly, it is building the machines that decide where to work next.