Shanghai chip designer Kiwimoore has confidentially filed for a Hong Kong initial public offering at a targeted valuation of about US$2 billion, according to two people familiar with the matter cited by Reuters reporting. The reported plan puts a company focused on AI-network interconnects among the latest Chinese technology firms seeking public-market capital as demand for computing clusters expands.
Kiwimoore is not primarily an AI model developer. Its role is the hardware layer that connects AI accelerators inside a computing cluster. Reuters says the company makes networking technology used to link accelerators. Its official website describes a full-stack portfolio for AI networking, including server-to-server, chip-to-chip, and die-to-die interconnect products intended for training, cloud inference, enterprise inference, and high-performance chips.
The distinction is important. A large AI cluster needs more than accelerators. It needs a fast, reliable way to move data among servers and chips. As models grow and as inference workloads serve more users, network performance can become a practical constraint. Kiwimoore’s reported IPO plan therefore offers a window into the less visible infrastructure layer beneath China’s AI buildout.
Reuters reports a confidential filing and a US$2 billion target valuation
Reuters reports that Kiwimoore filed confidentially for a Hong Kong IPO and is targeting a valuation of roughly US$2 billion. The news agency said the company did not respond to its request for comment. Because the filing is confidential and the company has not publicly confirmed the transaction, the reported valuation, timetable, and financing details should be treated as plans described by sources, not as finalized terms.
The sources told Reuters that Kiwimoore aims to list in the first half of 2027. They also said the company raised about 700 million yuan in a July funding round at an 8 billion yuan valuation and had selected CITIC Securities and ABC International to lead the offering. Reuters said the named banks did not respond to requests for comment.
These details are useful indicators of investor interest, but they do not establish that an IPO will take place. A confidential filing can be changed, delayed, or withdrawn. A target valuation is not the same as a market price. The report does not disclose Kiwimoore’s revenue, profitability, planned fundraise size, or customer concentration. Those gaps make it impossible to value the company from the available information alone.
Still, the reported filing is significant because it brings an AI-networking specialist into the same financing conversation as Chinese model, chip, and robotics companies. EastFrontier has reported on how China’s AI market is shifting toward a larger domestic share for local chipmakers. Kiwimoore occupies a related but different position: it is focused on the connections among components, not on the accelerators themselves.
AI clusters need interconnects at three different layers
Kiwimoore’s website divides its technology into three areas. Server-to-server interconnects address networking between machines in large-scale systems. Chip-to-chip interconnects are aimed at high-performance supernodes. Die-to-die interconnects use chiplet technology to connect components within a single chip package. Together, the company describes them as a full-stack approach to AI networking.
The company lists AI training clusters, cloud inference clusters, enterprise and edge inference clusters, and high-performance chips as application settings. Each setting presents a different technical problem. Training systems must coordinate a large amount of work across many accelerators. Cloud inference systems must serve many requests with predictable performance. Enterprise and edge systems often need to operate within tighter power, space, or local-data constraints.
These applications explain why networking has become an AI infrastructure issue. A faster accelerator does not automatically create a faster cluster if machines cannot exchange information efficiently. Large model workloads can require frequent communication among processors, and system designers must manage throughput, latency, and memory movement as well as raw computing capability.
The sources do not give Kiwimoore’s performance figures, specify its networking protocols, or name particular customers. They also do not show how its products compare with foreign or domestic rivals on standardized tests. The company’s website establishes its intended product categories, but not an independently verified performance lead. Any article about the IPO should retain that distinction.
China’s AI infrastructure strategy is often described through national goals for domestic chips. EastFrontier’s examination of China’s push toward 80 percent chip self-sufficiency by 2030 emphasized that the goal involves more than a substitute processor. It also requires the supporting stack that lets local hardware operate in useful, scalable systems. Interconnect technology is one of those supporting layers.
The IPO plan tests investor appetite for AI infrastructure specialists
A potential Hong Kong listing would test whether public investors are willing to assign a high valuation to a company supplying the connective tissue of AI clusters. The US$2 billion target reported by Reuters implies that Kiwimoore and its backers see AI-networking infrastructure as a category capable of attracting substantial capital even without the consumer visibility of a chatbot or a humanoid robot.
There are plausible reasons for that view. China is building more domestic computing capacity, and local chip systems need networking products tailored to their architecture and deployment needs. The company’s focus on training, cloud inference, enterprise inference, and chiplet connections positions it across several possible demand sources. But the sources do not show how much of that demand Kiwimoore currently serves or whether it has converted technical capability into recurring sales.
The company was founded in early 2021, Reuters reported. Its website calls it a full-stack AI networking solutions provider and frames its work around domestic-compute substitution. That positioning may be attractive while Chinese firms seek alternatives to foreign supply chains. It also exposes the company to a demanding market: customers will assess interoperability, reliability, software support, pricing, and the ability to keep pace with rapidly changing accelerator designs.
For investors, the eventual prospectus would be more revealing than the reported target valuation. It could show revenue growth, gross margins, customer concentration, research spending, supply-chain exposure, and the firm’s relationship with key hardware ecosystems. Until then, the headline should remain appropriately modest. Reuters has reported a confidential filing and a planned valuation, not a completed offering.
Kiwimoore’s reported move nevertheless highlights a broader feature of China’s AI competition. The opportunity is not limited to the companies that train models or fabricate chips. It also includes specialized suppliers that make large AI systems communicate efficiently. If Kiwimoore reaches the market on the timetable Reuters reported, its IPO will offer a public test of how investors value that infrastructure role. For now, it is an early sign that the network layer of China’s AI buildout is seeking its own capital-market moment.