China Telecom's Digital Human Empire From Cloud to Congress

China Telecom's digital human program has evolved from early experiments into one of the most technically ambitious virtual being ecosystems operated by any state-owned enterprise in China. The company's trajectory in this space reflects a convergence of cloud computing infrastructure, proprietary large language models, 3D reconstruction research, and consumer-facing applications that together position it as a significant player in the country's rapidly maturing digital human industry.

The foundational infrastructure supporting China Telecom's digital human ambitions rests on two pillars. The first is Tianyi Cloud (天翼云), the company's national cloud platform, which holds the top position in China's government cloud market and ranks among the top three in domestic public cloud IaaS. Generating 113.9 billion yuan in revenue in 2024, Tianyi Cloud provides the cloud rendering, computing, and AI model services that underpin the digital human ecosystem, including operating two 10,000-card GPU clusters under a distributed architecture designated "2+3+7+X." The second pillar is Xirang (息壤), an integrated intelligent computing platform whose research and development began in 2018, culminating in a formal release in 2022. Recognized by SASAC as one of the top ten super projects among centrally administered state-owned enterprises, Xirang enables unified scheduling of heterogeneous computing resources across regions and providers, with total capacity reaching 62 EFLOPS by the end of 2024, comprising 35 EFLOPS of self-built capacity and 27 EFLOPS contributed by fifty ecosystem partners. CEO Ke Ruiwen has designated Xirang as China Telecom's "First Technology," and the platform was upgraded to a new architecture in April 2025 with further enhancements planned for WAIC 2025.

On top of this infrastructure, China Telecom developed its Xingchen (星辰) family of large models under the TeleChat brand. Vice President Zhang Xin first publicly introduced the model on July 2, 2023, at the Global Digital Economy Conference AI Forum in Beijing, and the formal launch followed days later at WAIC 2023 in Shanghai. The Xingchen umbrella encompasses TeleChat for language, TeleSpeech for speech, TeleImage for multimodal processing, and Xinghe (星河) for visual tasks. Development is led by the company's dedicated AI entity, established in November 2023 with registered capital of three billion yuan and more than 800 research staff, under the scientific leadership of CTO Li Xuelong, a member of the European Academy of Sciences and the AAAI Executive Committee. The TeleAI research institute itself was formally unveiled at WAIC 2024, with Turing Award laureate John Hopcroft co-officiating the ceremony. The model has iterated rapidly, from a bilingual Chinese-English LLM at launch to the open-sourcing of TeleChat-7B in January 2024, making China Telecom the first centrally administered state-owned enterprise to release an open-source large language model. By September 2024, TeleChat2-115B became the first fully domestically trained open-source model exceeding 100 billion parameters, trained on Huawei Ascend Atlas 800T A2 hardware using ten trillion tokens. The December 2025 release, TeleChat3, introduced a mixture-of-experts architecture with approximately 4.7 billion active parameters and a chain-of-thought reasoning mode. Across the enterprise, the model has been deployed in 32 industry-specific configurations serving more than 7,600 clients, with applications spanning 18 customer service scenarios nationwide, 31-province network operations coverage, 95 percent government service scenario coverage, and a speech model supporting over 50 dialects that became the first independently filed speech model in China. China Telecom has characterized its ambition as becoming a "tens-of-billions-RMB-scale AI service provider."

In parallel with BAAI (Beijing Academy of Artificial Intelligence), the company jointly developed Tele-FLM-1T, described as the world's first one-trillion-parameter dense model. It also integrated DeepSeek's open-source model in February 2025, providing dedicated computing solutions for DeepSeek-R1, and the Ministry of Industry and Information Technology confirmed that China Telecom was leveraging digital humans' interactive capabilities to deliver new customer experiences.

China Telecom's 3D digital human reconstruction technology reached a notable milestone at the November 2023 Digital Technology Ecosystem Conference in Guangzhou. The company announced a fully automatic 3D reconstruction pipeline requiring only a few photographs to generate ultra-realistic digital humans, achieving minute-level processing with an average 3D vertex error of less than one millimeter. The system faithfully reproduces facial shape, wrinkles, skin texture, and fine details, with a semi-automatic topology binding workflow stable enough to support micro-expression-level motion down to the eye corners and mouth. Self-developed driving and rendering engines expand the range of emotional expression and movement. The company reported that this technology reduced digital human production time from one month to three days while cutting human labor steps by 80 percent. An earlier system, the 2022 digital human Xiaoyi (霄逸), had already demonstrated sub-millimeter error in expression reconstruction, and the 2023 advances represented a leap toward full automation. The production division, known as Dianxin Zhike (电信智科), employs approximately 600 staff, more than 70 percent in research and development, and won two championships at CVPR 2023 while placing in the top three across six additional competitions.

The company has assembled a substantial portfolio of named digital human characters, each tailored to a specific function. Weiyang (未央), a virtual male singer created by the Ai Music subsidiary, debuted in 2019. Xiaoyi (霄逸), the company's first virtual digital host, appeared at the 2022 Digital Technology Ecosystem Conference as a 3D ultra-realistic presenter. A separate Xiaoyi (筱翼) launched on the China Telecom mobile application in November 2022 as an AI digital employee providing 24/7 voice interaction with an elderly-friendly mode. Qianran (浅然), an ultra-realistic digital spokesperson for the music division featuring BSSRDF skin rendering and individually modeled hair, followed in February 2023. Shushu (数数) appeared alongside CCTV host Kang Hui at the 2023 Data Expo, while Xinyi (新翼) hosted the company's 2023 first-half earnings call, making it the first digital human to preside over a major state-owned enterprise earnings briefing. Xinyi was produced using 4D LightStage scanning, with a face comprising approximately 50,000 polygons and more than 600 facial bones driven by a bone-blendshape hybrid system. Na Jie (娜姐), a digital twin of Shanghai Telecom model worker Qiu Lina, hosted a labor recognition event in May 2024. The AI Customer Assistant, launched on World Telecom Day 2024, is an LLM-powered digital human in the company's mobile application capable of rendering two-dimensional video at a ratio of five seconds of processing per one second of output on a consumer GPU, with text-to-speech supporting six emotional states in Mandarin, Sichuan dialect, Cantonese, and English. Sulin (苏琳), a 3D intelligent digital panda, serves the TeleTrip cultural tourism platform in Aba Prefecture, Sichuan. Suixiaoxin (穗小信) functions as a policy interpretation digital human for the Guangzhou MIIT bureau, while Xiaoling (小翎) provides around-the-clock mental health counseling in Huzhou.

Consumer-facing digital human integration has extended into China Telecom's core telecommunications services. The Tianyi Communication Assistant introduced a digital human virtual assistant in late June 2024, and a major upgrade on September 10, 2024, added DIY avatar personalization across five visual styles: cartoon, pop, realistic, sketch, and exaggerated. Users upload personal photographs that AI renders into dynamic video personas appearing during calls, rolled out initially across 16 provinces. The complementary 5G Enhanced Calling service allows users to transmit digital human avatars with variable appearances during voice calls, with features including virtual backgrounds and gesture-triggered effects.

China Telecom has maintained a consistent presence at the World Internet Conference in Wuzhen, serving as both exhibitor and network infrastructure provider since the event's inception. In 2014, the company supplied the wireless network supporting 50,000 concurrent users. By 2019, it had deployed 36 5G macro stations and 243 indoor PRRUs to achieve full 5G coverage across the venue. At the 2023 edition, the company exhibited a 5G+VR motorcycle game experience at the Light of the Internet Expo, where attendees used VR headsets to enter a simulated racing environment powered by China Telecom's 5G platform, showcasing its high speed, massive connectivity, and low latency. At the 2025 summit, the digital human Xiaoyi (小翼) provided multilingual guide services across the exhibition, supported by a deployment of 717 personnel and 127 vehicles for network operations.

The company's partnership ecosystem extends well beyond any single technology provider. Huawei co-founded the China Telecom developer alliance in 2024 and provides Ascend computing infrastructure for TeleChat training. ZTE participates in both the developer alliance and 5G-A showcases. Volcengine (ByteDance's enterprise cloud), AsiaInfo, Yitu, and Zhipu AI are additional alliance and collaboration partners. Academic partnerships with Tsinghua University, Peking University, and Tongji University support AI talent cultivation through jointly built training centers.

On the standards and regulatory front, China Telecom has led the completion of 21 cloud-network autonomous operation standards at the ITU-T and developed information communication network intelligent operation management standards at CCSA. Its TeleAI semantic algorithm has been registered under the national deep synthesis algorithm filing system, and the Xingchen model series has completed the CAC's generative AI service filing. The Xingchen Speech Model became the first independently filed speech model in China. In the context of the draft Measures for the Administration of Digital Virtual Human Information Services published by the Cyberspace Administration of China on April 3, 2026, China Telecom's existing compliance infrastructure positions it to meet the regulation's requirements for mandatory labeling, consent for the use of personal likeness, algorithm filing, and sector-specific rules. Chairman Ke Ruiwen has stated that the company has built end-to-end AI security protection systems covering infrastructure, data, content, algorithms, and applications.

Looking forward, China Telecom's 2026 annual work conference adopted the theme of "fully embracing artificial intelligence," upgrading its corporate strategy to "Cloudification, Digital Transformation, AI for Good." The company's AI framework follows a "1+1+1+M+N" structure: one intelligent cloud foundation, one general model in Xingchen, one data foundation, M self-use models, and N industry models. At MWC 2026 in Barcelona, General Manager Liu Guiqing reported that network-oriented AI models have created digital employees that reduced maintenance staff repair trips by 35 percent, that AI-generated code now accounts for 40 percent of the codebase, and that overall research and development efficiency has increased by 20 percent. At the 2025 National People's Congress, Zhang Min, China Telecom's Hubei secretary and general manager, brought a personal digital human connected to large language models to interpret policy proposals, a gesture emblematic of the company's determination to embed digital human technology not merely in consumer entertainment but in the institutional fabric of Chinese governance and enterprise.

[Apr 2026]