PRESS RELEASE: ZTE Corporation has introduced AiCube, an innovative solution that redefines how telecom operators deploy and manage AI infrastructure.

This launch comes at a pivotal moment, as global data center capital expenditure surged by 38% in the first half of 2024, with telecom companies leading in the adoption of GenAI to optimize networks and enhance enterprise customer service.

As telcos accelerate their digital transformation, they face a range of challenges, from the high costs of public GenAI services to subscriber data privacy concerns and GDPR compliance. They also navigate geopolitical restrictions on GPU access, scalability limits in multi-GPU environments, and the considerable energy consumption of conventional AI data centers. ZTE’s AiCube directly addresses these industry-specific issues with a comprehensive, three-layered AI deployment framework.

Comprehensive Three-Layer Solution

At the hardware layer, AiCube features ZTE’s R6500 G5 8-GPU server, offering seamless compatibility with leading GPUs from NVIDIA, Intel and others. This infrastructure is further enhanced by ZTE KS20000 EB-level all-flash storage solution, providing high-performance parallel file storage, and the ZXR10 5900 RoCE DC switch, which supports the OLink open standard for connecting up to 10,000 GPUs. Additionally, AiCube’s advanced GPU/CPU cold plate liquid cooling system significantly reduces energy consumption, enabling telcos to optimize their existing infrastructure.

At the software layer, AiCube’s AI-Booster provides a one-click deployment system preloaded with zero-cost open-source LLMs, such as Llama, Qwen, Baichuan, and ChatGLM. Designed for smooth integration with AI frameworks such as PyTorch and TensorFlow, the software layer also includes performance optimization tools tailored for telecom-specific applications, such as chatbots, image and video generation, code generation, and medical image recognition.

At the service-layer, AiCube bridges the gap between AI technology and practical application by offering telecom-specific LLM services. These services include fine-tuning and integration for network operations, customer service automation, and predictive maintenance, along with ready-to-use GenAI application templates for common telecom use cases. AiCube demonstrates rapid scalability, supporting the pre-training of 13B LLM within 20 days and fine-tuning 130B LLM in just three days.

Proven Success Across Industries

AiCube is already making a significant impact in the telecommunications sector. At ZTE’s Nanjing R&D Center, the platform was used to build a code generation LLM for 20,000 software engineers, integrating hundreds of thousands of technical documents from 3GPP, IEFT, ETSI and mainstream open-source communities, with up to 100 billion tokens of communication cloud code. The system supports multiple programming languages, including Python, Go, and C/C++ achieving an impressive 80.5% HumanEval score, outperforming GPT-3.5. With 12,000 daily active users and a code adoption rate of 40-45%, the platform improved coding efficiency by 30% and boosted overall R&D productivity by 10%.

Innovation and Future Growth

Looking ahead, ZTE has outlined an ambitious roadmap for AiCube’s capabilities.  Upcoming innovations include heterogeneous parallel scheduling for ultra-large-scale cluster computing to handle growing network complexity. Additionally, the company plans to implement two-phase immersion liquid cooling technology, targeting an industry-leading Power Usage Effectiveness (PUE) of 1.1. ZTE also aims to expand its portfolio of telecom-specific GenAI application templates and ensure compatibility with emerging GPU technologies and open-source LLMs.

As of November 2024, AiCube has been deployed at over 100 sites, including both commercial and trial deployments. Its applications span multiple industries, including telecommunications, metallurgy, power, smart cities, healthcare and manufacturing, demonstrating the platform’s versatility and success across real-world use cases.

Enhanced Security and Control

Security remains a key pillar of the AiCube solution. By enabling on-premises deployment, AiCube ensures that telecom operators maintain complete control over their network and subscriber data. The platform’s adherence to open-source principles and standardized interfaces prevents vendor lock-in, giving operators the flexibility to modify or switch components as their needs evolve.

Paving the Way for AI-Driven Telecom Networks

ZTE’s AiCube represents a major step forward in AI deployment for telecom networks, offering a comprehensive and scalable solution that combines cutting-edge technology with practical usability. By addressing the unique challenges telcos face- such as data privacy, scalability, AiCube empowers operators to take control of their AI infrastructure. This solution paves the way for a more secure, efficient, and customizable future, enabling telecom companies to harness the power of artificial intelligence while retaining control over their data and operations.

With AiCube, ZTE is not just delivering an innovative product but setting a new standard for AI-driven telecom networks, providing operators with the tools to meet today’s challenges and thrive in tomorrow’s AI-powered landscape.