China Telecom's advantages in the field of artificial intelligence and the problems that need to be solved

China Telecom's advantages in the field of artificial intelligence

China Telecom has accurately grasped the development trend of artificial intelligence, continued to increase its efforts in scientific and technological innovation, and formed an overall layout of artificial intelligence of "1+1+1+M+N", forming unique advantages in computing power infrastructure, data, base large models and promoting the implementation of large model applications.

Advantage 1: Continue to promote the optimization and upgrading of computing power infrastructure and consolidate the foundation of Guoyun Intelligent Computing. First of all, the computing power resources are large in scale and the network coverage is wide. A "2+3+7+X" public intelligent computing cloud pool covering the whole country has been built, and a domestically-produced fully liquid-cooled single-cluster Wanka Intelligent Computing Center has been built in the Beijing-Tianjin-Hebei and Yangtze River Delta regions. The self-built intelligent computing capacity is 22EFLOPS, and the computing power of the access partners has reached 27EFLOPS. At the same time, China Telecom has built a flat, fully optical network, and intelligently operated intelligent and extremely fast all-optical network. The high-speed network covering urban and rural areas provides a low-latency and highly reliable foundation for AI computing power scheduling. Secondly , a distributed reasoning pool has been built. Construct an integrated intelligent computing service platform with "Xiran" as the core, integrating computing network scheduling, computing acceleration, and model training and promotion, and take the lead in completing the deep adaptation and optimization of domestic computing power and DeepSeek-R1/V3 series large models, and realize the implementation of full-stack domestic reasoning services. Finally, security capabilities are implanted into computing power operations, and self-developed vertical security models are used. Taking actual attack and defense scenarios as the entry point, they are deeply embedded in the computing power operation process to form five major capabilities, including digital landforms, intelligent penetration, and threat assessment.

Advantage 2: Gathering massive data resources to form a trusted and open full-stack data element capability system. On the one hand, accelerate the development and utilization of data resources. Build a "Xinghai" data resource map, integrate the average daily 1.6PB of telecommunications data, 27 industry data such as finance/Internet, and 33 provinces and cities/15 types of public data, and create an open and shared trusted data space through centralized lake entry and privacy interconnection. On the other hand, provide full-stack data element services through three platforms. Focusing on 10 core capabilities such as big data PaaS and AI multimodal governance, build three service platforms covering the entire process of data base, collection, cleaning, processing, labeling, security, and transactions. Among them, Xinghai Data Intelligent Middle Platform supports the scheduling of millions of mixed tasks and serves more than 150 government and enterprise customers. Lingze 2.0 Data Element Platform provides functions such as data rights confirmation and registration, data fusion calculation, and serves 6 provincial and 27 municipal data bureaus to expand the marketization of data elements. Galaxy Data Cross-border Platform has the full-link capability of cross-border data storage, calculation and transmission, and has supported pilot operations in many places.

Advantage 3: Independently develop the basic large model base and build an open source open ecosystem. First, the full stack independent research and development covers semantics, voice, vision, and multi-modality. TeleAI has completed the development of the first domestically produced Wanka Wancan large model in China. Secondly, it is the first among central enterprises to open source multi-size voice and semantic large models. Finally, it insists on open cooperation and continuously improves the service capabilities of Xingchen MaaS (Model as a Service), introducing 64 general and industry large models, and integrating 109 leading manufacturers' model capability tools.

Advantage 4: Relying on a massive user ecosystem, it has the ability to implement scenarios in all domains. On the one hand, it uses the advantages of "customers + channels" to drive application implementation. It has more than 400 million mobile users and more than 4.55 million Tianyi Cloud enterprise customers, providing a broad market space for the application and promotion of AI products; it continues to promote the upgrade of the "three transformations" of channels (chainization of own channels, intelligentization of online channels, and ecologicalization of social channels). The annual sales volume of the chain exceeds 10 million, the proportion of online AI services replacing traditional services exceeds 18%, and the number of social channels reaches 100,000. On the other hand, it actively expands the coverage of scenarios in all domains, and develops and applies more than ten large internal models for scenarios such as network operations and customer service to promote cost reduction, quality improvement, and efficiency improvement of enterprises. Through the self-developed AI energy-saving system, it is widely deployed in base stations, IT cloud hosts, and Class AB computer rooms, achieving an average annual intelligent power saving of about 800 million kWh. It has launched more than 50 large-scale scenario models for government affairs, industry, education and other fields, focusing on building more than 80 industry large-scale models and 20 industry intelligent bodies, providing integrated services of "computing power + platform + data + model + application", serving more than 10,000 customers; promoting the "AI + product" special action, and accelerating the large-scale promotion of more than ten products such as AI mobile phones, AI cloud computers, AI cloud disks, and AI cameras.

China Telecom accelerates the development of artificial intelligence and urgently needs to solve the problems

First, we need to further enhance the R&D capabilities of large-scale base model technology.

At present, the performance and influence of China Telecom's open source big model are weaker than those of leading companies. According to the IDC report, the application share of Xingchen big model in the domestic market ranks sixth (2024), belonging to the second echelon. At present, AI big model technology is advancing very rapidly. China Telecom especially needs to accelerate breakthroughs in computing power autonomy, data governance, product adaptation and algorithm innovation, further strengthen technology research and development and ecological construction, and improve the commercial application level of AI technology, so as to comprehensively narrow the gap with the world's leading artificial intelligence companies.

Second, we need to strengthen the construction of a team of top AI technology experts and AI application talents.

At present, China Telecom's AI talent pool cannot meet the needs of the 2025 National People's Congress and the Chinese People's Political Consultative Conference to continue to promote the coordinated development of "AI + industry" integrated applications, especially the lack of top AI experts, AI engineering talents and AI actual scenario application talents. China Telecom needs to continuously attract and cultivate AI talents through school-enterprise cooperation and flexible employment system.

Third, we need to further strengthen the construction of high-quality industry data sets to enable the industry’s intelligent upgrade.

The various industry data sets for the implementation of large models are affected by the industry's data privacy protection and data security compliance requirements and cannot meet the requirements for the application of general large models in vertical fields. China Telecom also needs to further strengthen cooperation with industry data partners to solve the problems of scattered data and high labeling costs, and jointly develop and build high-quality industry data sets.