During the World Conference on Artificial Intelligence (WAIC) 2025 in Shanghai, Zhaoxin highlighted a pair of internal processors of internal processor. These chips target opposite segments of the AI market, which is experiencing rapid expansion. The KAIXIAN KX-7000N is the first company class processor of the company to integrate a neural processing unit. On the other hand, the Kaisheng KH-50000 is a chip for data centers which allows Zhaoxin to offer servers with 96 cores.
The KX-7000N, which is in line with the Kaixian KX-7000aer series, adds an integrated NPU in a heterogeneous manner designed to accelerate the inference on the device. Zhaoxin considers this chip as the base of the PC of AI of the next generation, or “AIPC”, intended to treat locally large models of language, voice recognition and the generation of images. The demonstrations of the company in WAIC included AI personal assistants, smart office platforms and the creation of media in real time, all made offline on prototype systems developed with partners such as Lenovo Kaitian.
For corporate workloads, the new Kaisheng KH-50000 Triple the number of cores of its predecessor KH-40000 and extends the L3 cache to 384 MB, which corresponds to the capacity found on the rival parts AMD EPYC 9004. The processor also has 128 PCIE 5.0 tracks, a DDR5 ECC memory and an ZPI 5.0 Allows configurations to two or four sockets for a maximum of 384 cores per knot. Zhaoxin claims that the platform offers the calculation density and the high -speed E/S required for IA training clusters, high density servers and deployments of heterogeneous accelerators.
Beyond silicon, Zhaoxin’s stand presented “Edge-TO-Cloud” comprehensive solutions, including AI work stations, educational terminals and specific appraisses, ranging from automation of documents to medical imaging. By associating processors designed in the country with an open software ecosystem, the company says that it can offer safe and profitable alternatives in segments that traditionally depend on foreign processors.
These two launches underline the more general will of China to become independent in the field of advanced semiconductors. Although the detailed clock speeds and the architecture information remains confidential, the Zhaoxin roadmap now covers the AI workloads of customers, work stations and servers, which testifies to a concerted desire to compete more directly with the X86 established suppliers, both in terms of performance and set of functionalities.
