China Telecom's Digital Human Empire From Cloud to Congress

The distinguishing feature of China Telecom's digital human programme is not the realism of any single avatar but the completeness of the stack beneath it. The commercial platform giants that dominate most discussion of China's virtual beings enter the field chiefly as suppliers, selling models and production tooling to whoever will build on them. China Telecom occupies a different position entirely. It is a centrally administered state-owned carrier that has assembled every layer of the digital human industry within its own walls, from the physical compute at the bottom to the finished character at the top, and it deploys the result not toward an open consumer marketplace but inward, into the machinery of governance and enterprise. The trajectory implied by that ambition runs from cloud to congress: a firm that owns the infrastructure, trains the models, and manufactures the avatars, and whose most telling deployments end up presiding over earnings calls, interpreting government policy, and appearing on the floor of the National People's Congress. Read as a whole, the programme is less a bid for entertainment revenue than a demonstration of what vertical integration looks like when the integrator is an instrument of the state.

That completeness begins at the foundation, where two platforms carry the weight. Tianyi Cloud, China Telecom's national cloud, holds the leading position in the country's government cloud market and ranks among the top three domestic providers of public cloud infrastructure, generating revenue on the order of a hundred billion yuan in a single year and supplying the rendering and computing services on which everything above it depends, including two clusters of ten thousand accelerator cards each operated within a distributed national architecture. Beneath the branding sits an integrated intelligent computing platform called Xirang, whose development began in 2018 and which the company's chief executive has designated its "First Technology" — a label that signals priority rather than modesty. Xirang schedules heterogeneous computing resources across regions and providers, pooling capacity measured in tens of exaflops and drawing not only on the company's own build-out but on the contributions of a wide ecosystem of partners, and its recognition among the most significant projects undertaken by central state-owned enterprises marks the seriousness with which the state treats it. The point of describing this layer at all is that most digital human ventures rent it; China Telecom owns it, and the avatars it produces are an application built atop compute it controls end to end.

On that base the company constructed its Xingchen family of large models, released under the TeleChat brand and first shown publicly in the middle of 2023. The significance of the model programme lies less in any single benchmark than in the position it established. In early 2024 China Telecom open-sourced a compact version of TeleChat, becoming the first centrally administered state-owned enterprise to release an open-source large language model, and later that year it followed with a far larger system described as the first fully domestically trained open-source model to exceed a hundred billion parameters, built on domestic hardware rather than imported accelerators. That detail matters well beyond the specification sheet, because it places China Telecom's models inside the national project of technological self-sufficiency — a firm demonstrating that the entire pipeline, from silicon to trained weights, can be assembled without foreign dependency. A dedicated research entity, capitalised in the billions and staffed by hundreds of researchers under a scientifically credentialed chief technology officer, gives the effort institutional permanence rather than the character of a passing initiative, and its models now run in dozens of industry-specific configurations across customer service, network operations, and government service nationwide.

The visible face of all this is a body of reconstruction technology that turns compute and models into convincing human figures at industrial speed. China Telecom has developed an automatic pipeline that generates an ultra-realistic digital human from only a handful of photographs, reproducing facial shape, wrinkles, skin texture, and fine detail with vertex error below a millimetre, and compressing what once took a month of production into a matter of days while stripping most of the manual labour out of the process. Self-developed driving and rendering engines carry the result into motion, extending the range of expression down to micro-movements of the eyes and mouth, and the work sits inside a dedicated production division whose staff are overwhelmingly researchers rather than operators. The engineering ambition is real, but its purpose is the instructive part: the aim is not one flagship avatar but a factory capable of producing many, cheaply and quickly, so that digital humans can be spun up wherever the company wishes to place one. That industrialisation is what makes the next move possible, because a firm that can manufacture avatars at scale is free to deploy them not as novelties but as fixtures, and it is in the choice of where to fix them that the programme's intent becomes legible.

Where those figures have been placed is the heart of the matter. China Telecom's roster of named characters includes the expected consumer-facing presences — a virtual singer, mobile-application assistants offering round-the-clock voice interaction, tourism guides and mental-health counsellors serving particular localities — but the deployments that reveal the programme's direction are the institutional ones. In 2023 a digital human named Xinyi presided over the company's half-year earnings briefing, the first virtual figure to host a major state-owned enterprise's financial results, its face built from tens of thousands of polygons and hundreds of facial bones. Policy-interpretation avatars have since been embedded in provincial government bureaux, translating regulatory language for the public around the clock. And the technology has migrated into the core of the telecommunications business itself, appearing as personalisable avatars inside calling services that let ordinary users upload a photograph, have it rendered into a dynamic persona across several visual styles, and transmit that figure during an ordinary voice call. The pattern across these deployments is consistent: the digital human is not being sold as entertainment but installed as an interface between institutions and the people they address, a synthetic intermediary for the state's own communication.

That trajectory culminates in the setting the programme's framing invites. At the 2025 session of the National People's Congress, a China Telecom executive brought a personal digital human, connected to the company's large language models, to help interpret policy proposals — a virtual figure operating within the highest deliberative body of the Chinese state. It is a small gesture in scale and a large one in meaning, because it completes the arc the rest of the programme describes. A carrier that owns the cloud, trains the models, and manufactures the avatars has followed that capability all the way up from the data centre to the legislative chamber, and in doing so has signalled where it believes the technology belongs: not primarily in the consumer economy that preoccupies the platform giants, but woven into the institutional fabric of governance and enterprise, where a state-owned firm building for the state can put it directly to work.

The company's regulatory posture confirms the reading. As the Cyberspace Administration circulated its draft measures governing digital virtual human information services in 2026, carrying requirements for mandatory labelling, consent for the use of a person's likeness, and algorithm filing, China Telecom already sat inside the compliance architecture those rules describe, its models registered under the generative-AI regime and its speech system among the first independently filed in the country. For a private platform, regulation of this kind arrives as a constraint to be managed; for a state-owned enterprise whose digital humans are already migrating into government service, it arrives as something closer to confirmation that the ground it has been building on is the ground the state intends to standardise. China Telecom's corporate strategy, recently recast around the language of cloudification, digital transformation, and artificial intelligence for good, reads accordingly less as a market position than as an alignment — a firm that built the whole stack because the whole stack was always meant to serve an end larger than commerce, and whose journey from cloud to congress is the clearest available expression of that intent.