Q. Mr. Chen, how do you view the current trends in AI computing demand amid the rapid development of AI technology?
A. AI computing demand is undergoing a profound transformation. The evolution of large model technology is shifting AI computing needs from ultra-large-scale centralised training to more efficient distributed inference optimisation.
The launch of DeepSeek R1 has achieved stronger AI capabilities while significantly reducing costs. DeepSeek requires only thousands of H800 GPUs for training, which is an order of magnitude less than traditional ultra-large clusters. The emergence of DeepSeek has indeed optimised the computing demand for AI models, allowing a single model to operate efficiently with lower hardware resource consumption. This optimisation lowers the training and deployment costs of AI models, enabling more enterprises and developers to enter the AI field with lower barriers.
I's important to note that while the computing demand for individual models has decreased, the widespread application of AI technology continues to drive overall computing demand upward. In the future, we will see a shift in focus from extreme high-performance AI computing to distributed inference, with collaborative computing architectures of cloud, edge, and terminal becoming mainstream.
Q. What new opportunities do these trend changes bring to telecom operators and the computing infrastructure industry?
A. The success of DeepSeek indicates that the structure of the AI computing market is being reshaped. Operators and cloud computing companies need to provide more flexible and open computing infrastructure to meet AI demands across different scales and scenarios:
Individuals and small businesses will directly leverage DeepSeek's open API services to build their own AI-driven businesses.
Medium-sized enterprises will privatize the deployment of open-source DeepSeek-like models, leveraging their private data advantages for fine-tuning, which will drive demand for public cloud, private training resources, and integrated machines.
Large enterprises will simultaneously conduct training and fine-tuning of DeepSeek-like models and high-end inference deployment, fully utilising DeepSeek's performance to optimise their businesses, which will drive demand for private training pools and high-end inference pools.
In the face of such market trends, operators can build a new business model that integrates AI, cloud computing, and networking through various means such as MaaS API, IaaS cloud services, private cloud deployment, and AI-driven integrated machines to achieve a new round of growth!
Q. In light of these trends, what solutions do ZTE offer?
A. The world is entering an era of generalised intelligence, where AI computing has become the core driving force for technological breakthroughs and industrial transformation. ZTE adheres to the principle of open decoupling, enhancing computing with networks, and simultaneous training and inference, providing robust support for industries with its full-stack, all-scenario intelligent computing solutions.
In terms of open decoupling, we promote the decoupling of software from hardware, training and inference, and models, gathering global industrial advantages. We have built an open intelligent computing foundation that supports multiple manufacturers' GPUs, providing flexible computing solutions, and optimising AI computing efficiency through end-to-end intelligent hardware and AI-driven software collaboration.
In enhancing computing with networks, we work with industry partners to advance GPU open interconnection standards, relying on our deep expertise in computing and networking to achieve high-speed interconnection from chip-level, server-level, data center-level to cross-data center-level.
Regarding simultaneous training and inference, we have launched the AiCube intelligent computing integrated machine product to solve the 'last mile' problem of large model commercialisation. Our 'intelligent computing integrated machine' integrates various high-computing hardware bases, an easy-to-use training and inference platform, as well as mainstream AI-driven frameworks, models, and applications, boasting three core advantages: rapid delivery, convenient use, and comprehensive security assurance. These products have been successfully applied across multiple industries, including finance, healthcare, manufacturing, smart cities, and water conservancy.
ZTE's full-stack intelligent computing solutions enable operators to quickly build AI-driven computing platforms, providing customised services for users and seizing new opportunities in the AI-driven digital economy. We look forward to collaborating with global operators to promote the widespread application of AI technology.
RELATED STORIES
ZTE and TAWAL to drive digital expansion with new partnership
Indosat and ZTE bring digital connectivity to Indonesia’s remote regions
ZTE leads the way in simplified, efficient, and green network evolution
