At the recent MWC Barcelona 2026, Chen Xinyu, Vice President of ZTE, unveiled the company’s ambitious “AI Factory” full-stack solution, designed to tackle the increasing demands for computing power posed by large AI models. This innovative approach employs end-to-end co-design techniques to reshape traditional hardware infrastructures, aiming to provide global customers with an AI infrastructure that optimizes total cost of ownership (TCO) over its lifecycle.
As artificial intelligence applications continue to expand, the infrastructure demands associated with large models have reached unprecedented levels. Chen highlighted that conventional methods of hardware stacking are no longer sufficient to harmonize scale, efficiency, and cost. “We need comprehensive architectural reconstruction to maximize resource compute efficiency and accelerate the large-scale deployment of AI,” he stated. The “AI Factory” employs a full-stack co-design strategy, integrating advanced AI servers, hyper-nodes, a lossless network, the AI Booster operating system, and development platforms like the AI Agent Studio, all augmented by customized IDC infrastructure.
Central to this initiative is the concept of compute-network co-design, aimed at breaking the physical limits of computing density and scale. Chen explained two main breakthroughs: vertical scaling, which enhances density, and horizontal scaling, which expands the overall system capacity. The vertical scaling utilizes an OEX (Orthogonal Electrical eXchange) architecture within hyper-nodes, allowing for physical connections between compute and switching trays through vertical crosslinking. This design minimizes the reliance on high-speed cables, significantly boosting compute density while ensuring stable communication and eliminating risks associated with cable loosening. Enhanced by ZTE’s proprietary high-capacity switching chip, this system is capable of supporting terabyte-level bandwidth with latency under 100 nanoseconds, compatible with global standards.
On the horizontal scaling front, the “AI Factory” constructs clusters within single data centers through Scale-Out networks and further aggregates compute power across multiple data centers via Scale-Across networks. This approach facilitates the establishment of scalable AI factory foundations, allowing for extensive deployment of AI solutions.
The software-hardware co-design component of the “AI Factory” is equally critical in enhancing energy efficiency. Chen emphasized that to fully exploit hardware capabilities, a deeply integrated software system is essential. ZTE’s software stack acts as the operating system for intelligent computing resource pools, transforming disparate physical resources into cohesive compute services. Collaborating with major GPU manufacturers, ZTE incorporates techniques such as framework optimization and intelligent scheduling to significantly elevate performance metrics. Chen noted that these optimizations can lead to an increase in token throughput by five to twenty times, a substantial improvement in efficiency.
To address the complexities of large-scale cluster engineering, ZTE has introduced the “AI Factory Twin Platform.” This platform leverages digital twin technology to simulate various aspects of AI infrastructure, such as hardware selection and thermal management, enabling optimized performance and cost-effective design throughout the lifecycle of AI factories.
As a multifaceted engineering undertaking, the “AI Factory” encompasses various components, including chips, algorithms, hardware, clusters, software, and data centers. Chen emphasized ZTE’s commitment to applying its vast experience in communication systems and large-scale networking to advance AI infrastructure. “Extreme co-design is our core philosophy,” he concluded, underscoring ZTE’s intention to maintain an open and collaborative approach with global partners to foster a future-oriented intelligent computing ecosystem. This initiative aims not only to democratize AI technology but also to drive significant advances across various sectors of the economy.
See also
Nvidia Invests $4 Billion in Lumentum and Coherent to Boost AI Computing Power
India Boosts AI Sovereignty with $200B Investment in National Compute Capacity
Aikido Launches AO60DC Floating Datacenter Technology for GW-Scale AI Compute at Sea
Moms Urge Caution on AI in Schools: 70% Oppose Using AI for Student Data Collection
Nvidia Stock Dips to $179 Ahead of GTC 2026: Is This the Time to Buy?
