Zhaoxin unveils first NPU-equipped CPU and 96-core server chip at WAIC 2025

At the World Artificial Intelligence Conference (WAIC) 2025 in Shanghai, Zhaoxin highlighted a pair of in-house processors. These chips target opposite ends of the rapidly expanding AI market. The KaiXian KX-7000N is the company’s first client-class CPU to feature a built-in neural-processing unit. In contrast, the Kaisheng KH-50000 is a data center-grade chip that elevates Zhaoxin’s server offerings to 96 cores.

The KX-7000N, building on the KaiXian KX-7000 series, adds a heterogeneously integrated NPU designed to accelerate on-device inference. Zhaoxin positions the chip as a foundation for next-generation AI PCs, or "AIPCs," intended to process large language models, speech recognition, and image generation locally. Company demonstrations at WAIC included personal AI assistants, smart office platforms, and real-time media creation, all performed offline on prototype systems developed with partners such as Lenovo KaiTian.

For enterprise workloads, the new Kaisheng KH-50000 triples the core count of its KH-40000 predecessor and expands L3 cache to 384 MB—matching the capacity found on rival AMD Epyc 9004 parts. The processor also introduces 128 PCIe 5.0 lanes, 12-channel DDR5 ECC memory support, and a proprietary ZPI 5.0 interconnect that allows dual- and quad-socket configurations for up to 384 cores per node. Zhaoxin claims the platform offers the compute density and high-speed I/O required for AI training clusters, high-density servers, and heterogeneous accelerator deployments.

Beyond the headline silicon, Zhaoxin’s booth featured comprehensive edge-to-cloud solutions, including AI workstations, educational terminals, and industry-specific application stacks that spanned from document automation to medical imaging. By pairing domestically designed CPUs with an open software ecosystem, the company argues it can deliver secure, cost-effective alternatives in segments that have traditionally relied on foreign processors.

The twin launches underscore China’s broader push toward self-reliance in advanced semiconductors. While detailed clock speeds and architectural disclosures remain under wraps, Zhaoxin’s roadmap now spans client, workstation and server AI workloads—signalling a concerted bid to compete more directly with established x86 vendors in both performance and feature set.