Key Takeaways

  • China Telecom selected suppliers for 40,000 servers valued up to 11.55 billion yuan for its 2026 to 2027 build cycle.
  • Arm-based systems tied to Huawei’s Kunpeng ecosystem captured the largest share of the order.
  • The move reinforces China’s broader Xinchuang drive and signals how AI workloads are reshaping server architecture choices.

China Telecom’s latest procurement stands as a clear indicator of how China’s technology stack is being rebuilt from the inside out. The company has finalized orders for 40,000 high-performance servers, a deal that could reach 11.55 billion yuan, or about $1.7bn. What stands out is the architecture bias. Most units are based on Arm designs and linked to Huawei’s Kunpeng ecosystem.

The ordering split is straightforward. China Telecom requested 28,000 Arm-based systems and 12,000 C86 servers aligned with traditional x86-style workloads. Arm is the same architecture common in smartphones, although its use in servers has grown as power efficiency has become more important. In this case, all Arm units are tied to Kunpeng. Several domestic firms won supply slots, including Inspur, ZTE, H3C, and Lenovo, all of which maintain public ties to the Kunpeng ecosystem.

Huawei did not bid directly, yet Kunpeng-linked suppliers collectively captured about 71% of the order, roughly 8.16 billion yuan, according to Chinese trade press. This procurement model allows Huawei to remain central without being listed as the primary vendor—a dynamic that has surfaced in other tenders.

Analysts at IDC reported that China represented about 27% to 28% of global server revenue in 2023. Domestic players, especially Inspur and Huawei, have increased their share as geopolitical constraints shift the market. Forecasts suggest further architectural change. By 2027, more than 45% of worldwide servers could use Arm or RISC-style designs according to Gartner. This trend is tied to power needs in edge environments and the growing volume of distributed workloads.

The procurement aligns directly with Beijing’s Xinchuang campaign, which focuses on replacing foreign technologies in critical systems. It began in government computing and now reaches databases, cloud platforms, and hardware used in telecom and financial services. US export controls accelerated this shift and forced design changes across industries. A server order of this scale fits into a much broader localization cycle.

China Telecom, China Mobile, and China Unicom now regularly issue large server tenders. In April, China Mobile sought nearly 63,000 units. Last year, China Unicom called for 87,000. Most of those systems were built around architectures from Huawei or the Beijing-based Hygon. When state-owned carriers tilt in the same direction, suppliers follow. The alignment of these entities provides parallel ecosystems with the repeat, national-scale orders needed to mature.

AI workloads have become far more varied, driving demand in parallel. Analysts at McKinsey reported that enterprise AI deployments increasingly mix training and inference tasks with conventional compute. Telecommunications experts at IEEE have observed that scheduling heavy agent-style models often benefits from general-purpose CPUs rather than relying exclusively on GPUs. China’s build-out is moving beyond racks of graphics processors toward hybrid systems. Arm-based processors handle data movement and orchestration work that pairs naturally with GPU-accelerated tasks.

Carrier cloud environments often align with ITU-T guidance for network function virtualization and distributed data centers. Deployments at this scale usually reference ITU-T Y.3500 series frameworks. Information security requirements commonly follow ISO/IEC 27001. These guidelines shape how operators procure and integrate new compute hardware, establishing baselines for security and operability.

Even with rapid domestic progress, technical bottlenecks remain. High-end AI training relies on chips that China cannot yet manufacture at scale. Moreover, Kunpeng’s dependence on Arm means that licensing rules remain relevant. That dependence is one reason designers continue to explore RISC-V. The open nature of that architecture reduces geopolitical risk, leading suppliers to treat it as a long-term option.

Analysts at Omdia project that operator capital expenditure for cloud and IT infrastructure in China will grow at a mid-single-digit compound rate through 2027. Traditional network investments are flattening, while computing and storage are taking a larger share. Large server contracts therefore hold strategic weight because they influence both vendor ecosystems and chip design priorities.

As these systems integrate with telco cloud platforms, software vendors take notice. China Telecom has been investing in cloud services for years, and its work on distributed architectures aligns with broader efforts by Chinese carriers to support edge AI services. Application portability and optimization are emerging as new areas of competition when carriers adopt Arm-based systems at this scale.

Global server architecture trends are shifting concurrently. Power efficiency, thermal limits, and workload diversity are shaping design choices in data centers worldwide. Meanwhile, Arm’s presence in hyperscale environments has expanded. The procurement demonstrates how quickly structural changes materialize when a large national buyer moves in one direction.

A single order does not reshape the market, but repetition does. China Telecom’s selection is part of a multi-year pattern among major carriers. Huawei’s Kunpeng ecosystem, strengthened by the indirect bidding structure, secures the bulk of this volume. As more AI infrastructure relies on mixed CPU-GPU configurations, Arm-based processors will likely continue gaining ground. Whether this momentum establishes a fully independent domestic computing stack remains an open question, but the trajectory is now visible.