What Drives China's 2024‑2026 CPU Landscape? A Deep Dive into X86, ARM, LoongArch and More

In 2023 China entered a new development cycle for the information‑technology innovation (信创) industry. The hardware market is expected to reach 7,889.5 billion CNY by 2026, driven by expanding civilian applications and a shift in procurement focus toward CPU performance, ecosystem compatibility, migration cost, and supply‑chain robustness.

1. CPU Instruction‑Set Foundations

CPU cores are built on instruction sets and micro‑architectures. Instruction sets fall into two categories:

CISC (Complex Instruction Set Computing) : A single instruction performs a complex operation. Examples include the x86 architecture, dominant in desktops, PCs and servers, with rich software and hardware ecosystems.

RISC (Reduced Instruction Set Computing) : Each instruction performs a simple action; multiple instructions combine for complex tasks. Representative architectures are ARM, MIPS, ALPHA and RISC‑V, widely used in mobile terminals and specialized devices.

2. Mainstream Instruction‑Set Comparison

Current dominant instruction‑set architectures are x86 and ARM. Market‑share data shows:

x86 offers high performance and strong compatibility, commanding over 90 % of the global server market and a similarly high share in China’s server sector.

ARM provides low power consumption and high integration, leading the mobile terminal market.

3. Licensing Models

x86 instruction sets are cross‑licensed between AMD and Intel, allowing companies to design CPUs without IP infringement concerns. ARM, however, licenses only specific versions (e.g., ARMv8, ARMv9) and prohibits downstream modification, limiting the ability of licensees to extend the architecture without additional agreements.

4. Domestic CPU Development Landscape

Historically, China relied heavily on foreign IT products (Intel, Microsoft, Oracle, Cisco). Security incidents and the need for controllable technology have spurred the development of a domestic ecosystem centered on CPUs and operating systems.

Two major ecosystem tracks have emerged:

Wintel – based on x86.

AA – based on ARM.

The six leading domestic CPU vendors and their instruction‑set origins are:

Shenwei (SW_64) – uses the proprietary SW_64 set.

Loongson (LoongArch) – develops the LoongArch set.

Zhaoxin – adopts the x86 instruction set.

Hygon – also based on x86.

Huawei Kunpeng – licensed ARMv8.

Phytium – licensed ARMv8.

Key observations:

x86‑based vendors (Zhaoxin, Hygon) benefit from a mature software ecosystem but must navigate cross‑patent licensing.

ARM‑based vendors (Kunpeng, Phytium) enjoy strong mobile‑terminal ecosystems but face licensing restrictions that hinder adoption of newer ARM versions (e.g., ARMv9).

Loongson and Shenwei, with self‑developed instruction sets, achieve high autonomy and control, yet their software ecosystems are still limited.

5. Comparative Analysis of the Six Domestic CPU Companies

All six have achieved national‑level 信创 certification and claim autonomous controllability. From an iteration capability perspective:

Loongson and Shenwei possess independent technology stacks (LoongArch, SW_64) and are not constrained by external IP.

Zhaoxin and Hygon rely on x86 patents; however, they can extend the instruction set internally and continue product innovation.

Kunpeng and Phytium are limited by ARMv8 licensing, which restricts the incorporation of newer ARMv9 features and may impede future product updates.

Market positioning:

Loongson focuses on servers and desktops, with emerging trials in critical‑information‑infrastructure projects.

Hygon and Zhaoxin target servers, desktops and embedded markets.

Kunpeng and Phytium serve servers, desktops and embedded devices, but their ARM‑based designs face ecosystem compatibility challenges.

Shenwei concentrates on high‑performance servers and compute‑intensive infrastructure.

6. Conclusions

Overall, x86‑based domestic CPUs (especially those from Hygon and Zhaoxin) enjoy the strongest market acceptance due to a mature ecosystem and broad software compatibility. ARM‑based solutions (Kunpeng, Phytium) have clear advantages in mobile and low‑power domains but are constrained by licensing limitations that affect long‑term evolution. Self‑developed instruction sets (LoongArch, SW_64) provide the highest degree of autonomy and security, yet their limited software ecosystems pose adoption challenges.

Strategic recommendation: leverage the strengths of each architecture—x86 for server‑grade performance, ARM for power‑efficient workloads, and indigenous instruction sets for security‑critical applications—while fostering ecosystem development to reduce compatibility gaps.