36Kr Exclusive | ZJU-Founded Desktop CNC Team Secures Nearly RMB 100 Million Angel Round from SenseTime Guoxiang, FirstForm Tech and Others, Aims to Lower Manufacturing Barriers with AI Technology
Author | Zhang Ziyi
Editor | Yuan Silai
Hard Krypton learned that desktop-level CNC (Computer Numerical Control) enterprise "Qisu Technology" has recently completed an angel round financing of nearly 100 million RMB. This round of financing was jointly invested by SenseTime Guoxiang, Shouxing Technology, New World, and Qiji Chuangtan. The funds from this round will be mainly used for the R&D, mass production of desktop-level CNC software and hardware products, as well as subsequent marketing.
After consumer-grade 3D printers and laser engravers have completed mass market education and spawned multiple leading enterprises such as Bambu Lab and xTool, CNC products that can achieve "strong 3D processing from soft materials to hard materials" are regarded as the next trillion-dollar blue ocean in the consumer-grade manufacturing sector.
As early as 2018, when the industry was generally focusing on the 3D printing track, the Qisu Technology team had already set their sights on the CNC field. In the eyes of Xia Nan, the founder, CNC is the most basic and versatile processing method in the manufacturing industry, covering more scenarios than 3D printing and possessing a trillion-dollar industrial foundation. If AI can be used to reconstruct the manufacturing paradigm of CNC and lower its barriers, it will further expand the application scenarios of CNC, and even extend it to the desktops of ordinary users. This will transform the entire society's production model and unleash enormous incremental value.
At present, several consumer-grade CNC startups have emerged in the market, and capital is also pouring into this sector. A common problem faced by all enterprises is: how to lower the barriers to CNC usage. This is a difficult problem that the industrial sector has spent decades trying to solve without success.
Specifically, the first challenge lies in the extreme complexity of using CAM software. CAM, or Computer Aided Manufacturing, is the core link that converts 3D models into processing instructions recognizable by CNC machines. It directly determines the processing accuracy, efficiency, and feasibility, making it the part with the highest technical threshold in the CNC operation chain. Users need to configure dozens or even hundreds of process parameters to complete the process design of a single part.
The second challenge is that CNC processing relies heavily on accumulated experience. Skilled operators need to consider numerous complex variables in their minds: how to clamp the workpiece, which tool to use, how to configure processing parameters, and so on — all of which are highly dependent on personal practical experience.
Xia Nan told Hard Krypton: "It takes 3 to 5 years to train a skilled CNC operator in a factory. Even for engineers who graduated from top-tier universities in China (the 985 Project institutions), mastering CNC requires at least half a year of full-time dedication."
In Xia Nan's view, the best solution to lower CNC barriers is AI CAM — integrating complex process knowledge into the system, equipping every machine tool with an AI-powered veteran operator.
Since 2018, Qisu Technology has been conducting R&D on AI CAM. Initially, the biggest problem they encountered was the lack of data. In the traditional CNC industry, processing process knowledge is mostly passed down through verbal instruction and personal mentorship, failing to form precipitatable digital assets that can be quickly transferred. In addition, the complexity of real processing scenarios far exceeds expectations. Even veteran operators with years of experience often encounter brand-new cases they have never faced before. This constitutes the core R&D difficulty of AI CAM software: how to digitize scattered, experience-dependent process knowledge and convert it into a precipitatable, reusable reasoning and decision-making model. Even leading overseas industrial software manufacturers have not yet developed a fully mature fully automated solution to this day.
To master process knowledge, the Qisu team went deep into frontline operations and accumulated thousands of hours of practical experience. To accumulate process data and validate it in real scenarios, Qisu Technology built its own flexible factory based on AI CAM, constructing a data closed-loop by accepting and fulfilling real production orders to continuously iterate its algorithm models.
Hard Krypton learned that Qisu Technology's AI CAM system demonstrates its core capabilities mainly in two aspects:
First, it simplifies the operation process and eliminates professional learning barriers. Traditional CAM follows the logic of "humans adapting to the software," requiring users to possess systematic processing process knowledge and manually configure hundreds of parameters. Qisu Technology reverses the interaction logic to "software adapting to humans" — users only need to import the model, and the software condenses complex CNC programming into three simple steps: process generation, preview, and processing. Users do not need to master professional processing knowledge or programming skills, as the software will guide them on what to do.
Second, it offers high scenario coverage with comprehensive supporting data. Leveraging hundreds of thousands of real processing model data accumulated by the team, the system's algorithms continuously learn and upgrade. With the underlying innovative AI CAM architecture and decision-making model, the entire generation process requires no human intervention. The system's data covers a large number of "corner cases," which can support the rapid generation of process plans and improve user experience.
Based on its years of self-developed AI CAM system, Qisu Technology plans to officially launch its first desktop-level 5-axis CNC in Q4 this year. The product will come standard with automatic tool change (tool magazine), automatic tool setting, a fully enclosed dust-proof and silent cabin, and a 1500W high-power spindle. Supported by the AI CAM system, it will achieve "end-to-end" ease of use.
In terms of team background, Qisu Technology's core team members come from the School of Computer Science and School of Mechanical Engineering at Zhejiang University. All members are hardcore makers with over 10 years of experience, having dabbled in various hardware fields such as exoskeletons and 3D printers, and possessing mass production experience of hundreds of thousands of units.
With the growing demand for personalized manufacturing and rapid iteration, the application boundaries of desktop-level manufacturing equipment continue to expand, and the desktop-level CNC track will also welcome more market participants and technological iterations.
CEO Dialogue:
Hard Krypton: Back in 2018, why did you choose to enter the CNC track?
Xia Nan: Our judgment at the time was that the digital intelligence and popularization of the manufacturing industry is a long-term, definite trend. CNC is the most basic and versatile processing method in the manufacturing industry, corresponding to a trillion-dollar industrial foundation. The manufacturing process of almost all products directly or indirectly involves CNC processing. If we can reconstruct the manufacturing paradigm of CNC and extend industrial-grade processing capabilities to a wider range of scenarios, even desktop environments, it will unleash enormous incremental value.
Hard Krypton: Your team has been working on AI+CNC related projects all along. What is the biggest difficulty in developing self-developed AI CAM technologies?
Xia Nan: The core difficulty lies in the complexity of real manufacturing scenarios. In the traditional manufacturing industry, processing process knowledge is mostly passed down through verbal instruction and personal mentorship, failing to form precipitatable digital assets that can be quickly transferred. Moreover, there are too many variables in real processing — even veteran operators will encounter brand-new cases they have never seen before. Converting these scattered, experience-dependent process knowledge into precipitatable, reusable decision-making models is extremely challenging. Leading overseas industrial software manufacturers that have been deeply involved in this field for decades still have not developed a fully mature fully automated solution.
This field requires genuine long-term commitment. AI CAM can only mature through extensive validation and iteration in real scenarios, and this process cannot be rushed. For example, one of our algorithms was written in 2021, and its stability did not meet the requirements for large-scale implementation until the end of 2024. The entire R&D process relies on long-term technological accumulation and scenario validation.