Zhaoxin, the Chinese processor maker that designs its own chips, presented its self-developed compute strategy for edge AI (AI processed on end devices) at the Edge AI Summit, part of the 10th Jiwei Conference 2026, which opened in Shanghai's Zhangjiang area on May 28[1]. Built around its in-house ZX86 instruction set, the lineup spans from server processors to AI PCs and is positioned as a compute foundation that connects cloud, edge, and device[1]. The move reflects the company's aim to capture the fast-growing edge AI market with domestic processors that do not depend on overseas vendors.

Edge AI Approaches Its Breakout Year, with Domestic Compute as the Base

Zhaoxin frames 2026 as a turning point for the mainstream adoption of edge AI, noting that AI processing is shifting in a big way from the cloud down to devices such as AI PCs[1]. According to an industry forecast presented at the summit, the edge AI market will grow from 321.9 billion yuan (about 7.6 trillion yen) in 2025 to 1.22 trillion yuan (about 29 trillion yen) in 2029, a compound annual growth rate of 40 percent[1].

※1 yuan = about 23.6 yen (as of June 9, 2026)

Edge AI offers low latency, the ability to keep data on the device for better privacy, and resilience to network conditions, and it is spreading across government and corporate offices, education, factory visual inspection, and more[1]. Zhaoxin argues that a domestic, self-controlled compute base is becoming increasingly important as the foundation for these use cases[1].

A Gradient of AI Compute Built Around ZX86

At the core of its technology is the ZX86 instruction set that Zhaoxin has defined and extended in-house[1]. The company says it follows a path that is self-controlled but not closed, and compatible with the mainstream but not dependent on it, advancing its own designs while maintaining compatibility with mainstream environments including x86[1].

Around ZX86, the company offers a product lineup tiered by use case[1]. The server-class KaiSheng KH-50000 provides 96 cores per socket and targets large language model training and high-density inference[1]. The KaiSheng KH-40000 offers up to 32 cores and, in dual configurations, enables cost-conscious servers and workstations aimed mainly at high-density inference at and near the edge[1]. The KaiXian KX-7000 series for PCs and embedded devices supports lightweight AI inference in offices, education, and industrial control on AI PCs and edge equipment[1].

From Servers to AI PCs: A Lineup Matched to Each Device

Working with industry partners, Zhaoxin assembled finished products that use its CPUs, including AI servers, AI workstations, and AI PCs[1]. AI servers built on the KH-50000 combine compute on the scale of 192 cores with multiple PCIe 5.0 slots, pairing with GPUs and AI accelerators from major domestic brands to support running large models and AI agents locally[1]. AI inference workstations built around the KH-40000 were shown for finely divided uses such as parallel computing, industrial image processing, low-power local inference, medical image analysis, and scientific computing[1].

On the device side, Lenovo's Kaitian worked with Zhaoxin to build what it describes as the first AI PC for the domestic environment using the KaiXian KX-7000[1]. In addition, partners such as iFLYTEK and Seewo offer AI PCs using KaiXian processors to support lightweight AI tasks in government and corporate offices and in smart education[1].

A Widening Ecosystem and Deployment Cases

Zhaoxin's in-house CPUs support the x86 instruction set and run mainstream AI frameworks out of the box, and they are compatible with domestic operating systems such as Galaxy Kylin and UOS[1]. The company is also working on compatibility with domestic GPUs and AI accelerator cards, databases, cloud platforms, large language models, and AI agents to lower the cost and effort of adopting edge AI[1]. The ZX86 compute platform is expanding support for open-source and domestic AI agents as well[1].

As a deployment example in the government and corporate office space, an all-in-one AI terminal involving People's Daily Online was introduced[1]. Built on a four-card AI workstation on the ZX86 platform, it combines a dedicated large language model with a large volume of specialized corpora to handle drafting official documents, automatically proofreading content, and centrally managing materials on a single machine[1]. In education, Lenovo Kaitian AI PCs and Seewo's education terminals built on the KX-7000 deploy lightweight education models on the device to support lesson preparation, in-class interaction, and analysis of learning progress[1]. A case of an AI practice classroom assembled by Zhaoxin in a fully domestic environment at a middle school in Beijing was also cited[1].

Summary

Rather than the specifications of a single new chip, this announcement laid out Zhaoxin's strategy of using the in-house ZX86 instruction set to tie together everything from servers to AI PCs into one compute foundation, raising the profile of domestic players in the fast-growing edge AI market. By presenting the full picture, including compatibility with domestic operating systems and AI agents and deployment cases in government and education, the company showed where it stands today.

出典:https://www.zhaoxin.com/news_view.aspx?nid=2&typeid=289&id=2044