🐯虎嗅•Freshcollected in 35m
China Mobile Builds an AI Computing Network
#ai-data-centers#gpu-orchestrationchina-mobile-ai-computing-networkchina-mobileaidcgpunputoken-factory
💡China Mobile is applying telecom-scale orchestration to GPUs, Tokens, and AI data centers.
⚡ 30-Second TL;DR
What Changed
H1 2026 computing-service revenue reached 529 billion yuan, up 14%, while telecom-service revenue fell 5.7%.
Why It Matters
China Mobile is positioning itself as an infrastructure and orchestration provider for enterprise AI, not merely as a connectivity carrier. If utilization and cross-region scheduling remain high, its network model could reduce fragmentation in China’s AI-compute market; however, GPU depreciation, chip heterogeneity, and customer demand remain major risks.
What To Do Next
Benchmark your inference workloads on China Mobile’s intelligent-computing service or equivalent multi-region scheduler, measuring latency, utilization, cost per token, and failover behavior before committing workloads.
Who should care:Developers & AI Engineers
Key Points
- •H1 2026 computing-service revenue reached 529 billion yuan, up 14%, while telecom-service revenue fell 5.7%.
- •2026 planned investment in computing networks rose 62.4% to about 37.8 billion yuan, while communications-network investment fell 20.3%.
- •Intelligent-computing service revenue reached 5.3 billion yuan, growing 130.1% year over year.
- •China Mobile reported 112.7 EFLOPS of intelligent-computing capacity, over 90% utilization, and average node latency below 12 milliseconds.
- •The company is using a '4+N+31+X' infrastructure structure and establishing Token and computing offices for unified orchestration.
🧠 Deep Insight
AI-generated analysis for this event.
🔑 Enhanced Key Takeaways
- •China Mobile has integrated its 'Jiutian' (Nine Heavens) AI model family into this computing network, specifically leveraging the 'Jiutian-Zhongqing' base model for enterprise-grade intelligent computing tasks.
- •The '4+N+31+X' architecture refers to 4 major regional intelligent computing hubs, N provincial-level centers, 31 core nodes, and X edge computing points, designed to minimize data transit for real-time AI inference.
- •China Mobile is actively deploying a proprietary 'Computing Power Network' (CFN) operating system that utilizes blockchain technology to ensure secure, transparent settlement of computing resources between different data centers.
- •The company has initiated a 'Computing Power Express' (Suanli Kuaiche) program to provide dedicated, low-latency private network channels for large model training, specifically targeting domestic AI startups and research institutions.
- •To address GPU supply constraints, China Mobile has developed a heterogeneous computing scheduling platform capable of pooling domestic AI chips (such as Huawei Ascend) alongside international hardware to maintain service continuity.
📊 Competitor Analysis▸ Show
| Feature | China Mobile (AIDC) | China Telecom (Cloud-Network) | China Unicom (CUBE-Net) |
|---|---|---|---|
| Primary Focus | Massive-scale intelligent computing | Hybrid cloud & government services | Industrial internet & edge AI |
| GPU Strategy | Heterogeneous pooling (Domestic/Intl) | Tianyi Cloud (Proprietary focus) | Collaborative ecosystem |
| Network Edge | 12ms latency target | 15-20ms latency target | 15-20ms latency target |
| Market Positioning | Infrastructure-as-a-Service (IaaS) | Platform-as-a-Service (PaaS) | Industry-specific solutions |
🛠️ Technical Deep Dive
- Architecture: Utilizes a hierarchical '4+N+31+X' topology to distribute AI workloads across national, regional, and edge tiers.
- Orchestration: Implements a unified 'Computing Power Network' (CFN) brain that treats GPU cycles as a tradable commodity, similar to bandwidth.
- Latency Optimization: Employs SRv6 (Segment Routing over IPv6) and all-optical switching technologies to maintain sub-12ms node-to-node latency.
- Heterogeneous Support: The scheduling layer supports abstraction of diverse hardware backends, allowing seamless task migration between different chip architectures.
- Security: Integrates confidential computing (TEE) at the node level to protect model weights and training data during cross-region scheduling.
🔮 Future ImplicationsAI analysis grounded in cited sources
China Mobile will surpass traditional telecom revenue with computing services by 2028.
The current growth trajectory of 130% in intelligent computing revenue against the decline in traditional telecom services suggests a structural pivot in the company's core business model.
The '4+N+31+X' network will become the primary backbone for China's sovereign AI development.
By standardizing computing as a utility, China Mobile is positioning its infrastructure as the mandatory layer for domestic AI companies to comply with national data security and localization requirements.
⏳ Timeline
2021-10
China Mobile officially proposes the 'Computing Power Network' (CFN) strategy.
2023-05
Launch of the 'Jiutian' AI model platform for industry-specific applications.
2024-02
Completion of the first phase of the national intelligent computing center cluster.
2025-06
Integration of Token-based billing systems for computing resource scheduling.
2026-03
Announcement of the '4+N+31+X' infrastructure upgrade to support 100+ EFLOPS capacity.
📰
Weekly AI Recap
Read this week's curated digest of top AI events →
👉Related Updates
AI-curated news aggregator. All content rights belong to original publishers.
Original source: 虎嗅 ↗