The Four Commercial Giants Powering China's Digital Human Industry

The commercial core of China's digital human industry is held not by the telecom operators that deploy avatars inside their service networks, nor by the content platforms that host and monetise them, but by four technology firms whose models, platforms, and cloud infrastructure supply the underlying capability on which most deployments across the economy ultimately rest. Baidu, Alibaba, Tencent, and ByteDance occupy that core together, and yet the most revealing fact about their position is how little it resembles dominance. IDC's mid-2025 measurement placed the market leader, Baidu, at 9.8 percent, a fraction of a point ahead of Huawei Cloud and trailed by a field distributed thinly below it — which is to say that more than ninety percent of the platform market lies in other hands, spread across specialist vendors, state-owned enterprises building proprietary capability, and a long tail of smaller providers. What holds these four firms together is therefore not a shared grip on the market but a shared function as its suppliers, and what separates them is more instructive still: each has entered the same field through a fundamentally different wager about what a digital human business actually is. Their divergence, rather than any rivalry between them, is the structure worth reading.

Baidu has made the most literal wager of the four, building the most fully realised digital human platform in China and earning the corresponding leadership its rivals cannot claim. Its Xi Ling platform, unveiled in late 2021 as a one-stop production and business-configuration service, grew out of commercial work that predated it — a digital bank employee named Xiao Pu, designated the banking industry's first digital staff member in 2019 — and matured into a system that by 2024 could produce a hyper-realistic avatar for a fraction of its former cost. The bet's clearest vindication came through the e-commerce platform Huibo Xing, which by late 2025 had generated hundreds of thousands of avatars and captured the leading share of the livestream digital human category. Its defining moment arrived in June 2025, when a digital twin of the entrepreneur Luo Yonghao livestreamed for more than six hours and generated over fifty-five million yuan before an audience of thirteen million, the avatar calling its own knowledge base thousands of times and coordinating its own gestures in real time — a performance Chinese media christened the livestream industry's DeepSeek moment. Underneath sat the conversational layer that distinguishes Baidu's figures from pipeline-only competitors: its ERNIE model family, which the company made free to all users in 2025 under competitive pressure and which anchors an assistant with a user base numbered in the hundreds of millions. Baidu's is the strategy of ownership from model to avatar to marketplace.

ByteDance has wagered instead on the foundations rather than the finished figure, treating the digital human as one application of a broader investment in multimodal generative AI. Its enterprise cloud division, Volcengine, offers a full digital human product line, but the company's real influence has been exercised through the models beneath it. Its release of OmniHuman in early 2025, generating lifelike video from a single image and motion input, and its steady advances in speech synthesis gave the wider ecosystem tools that competitors adopted rather than merely admired. That same continuity runs upward from the research bench to the consumer surface: the same investment that produces enterprise avatars also powers Doubao, the assistant that by early 2026 had become the country's most-used consumer AI product by a wide margin, and a suite of creative tools that let ordinary Douyin creators generate digital versions of themselves. More consequential still was a pricing decision: in May 2024 Volcengine cut its enterprise model prices to figures roughly ninety-nine percent below the prevailing average, triggering the LLM price war whose effects rippled through digital human economics across the sector. The counter-signal came the same year, when ByteDance sold the virtual idol group A-SOUL — a genuine cultural phenomenon, and one whose backstage disputes over performer treatment had already bruised it — to an entertainment company, narrowing its investment away from owned virtual-idol IP and toward the infrastructure it preferred to sell. Where Baidu builds the whole stack, ByteDance builds the layer everyone else builds on, and sets its price.

Alibaba's contribution is the most diffuse and, in one respect, the most consequential for the field's health: it has distributed capability outward rather than concentrating it. Its Tongyi Qianwen family, released internationally as Qwen, has been opened at successive scales under a permissive licence, and by early 2026 had become the world's most downloaded model family, ahead of Meta's Llama, with tens of thousands of community-built derivatives — an open-weight strategy that seeded countless digital human builders who never touched an Alibaba product directly. Its speech system CosyVoice was released fully open-source on comparable terms, and its research arm contributed tooling aimed squarely at digital human production — EMO, which animates a single photograph into a singing or talking figure without any three-dimensional modelling, alongside a wider stack of talking-head and character-animation systems folded into the same open research corpus. Against that quiet infrastructural generosity sit a small number of highly visible set-pieces, most of them routed through the affiliate Ant Group: the digital torch-bearer for the 2023 Hangzhou Asian Games, assembled from three-dimensional avatars of more than a hundred million participants and used to co-light the main flame in a first for the Games' history. Alibaba's commerce platforms have meanwhile formalised digital human governance through livestream registration and labelling requirements, whilst its earlier flagship metahuman, AYAYI, demonstrated the strategy's limit — a lavish brand career curtailed as per-second production costs made her impossible to scale. Alibaba supplies the substrate and stages the spectacle, and lets ordinary commercial use grow within its marketplaces.

Tencent has made the most distributed wager of all, favouring cultural-institutional partnership and indirect equity over dominant platform ownership. Its most distinctive footprint lies in heritage: a digital cultural ambassador for the Dunhuang caves, and the museum guides launched at the National Museum of China, drawing on a knowledge base of well over a million artefacts. It has taken a substantial stake in Silicon Intelligence, the leading digital human agent firm by revenue, giving it a large position in a market it does not directly operate, and has continued to cultivate proprietary virtual-idol IP where ByteDance retreated from it. Yet Tencent also embodies the sharpest internal contradiction in the sector. In June 2024, WeChat Video Channels banned digital human livestreaming outright, classifying both AI-driven and human-driven avatars as non-authentic content and excluding them from its tipping, commerce, and payment functions — the first major Chinese platform to do so, and a decision that walled the company's own enterprise digital human services off from its dominant consumer surface even as Tencent helped draft financial-sector standards for the very technology it had just restricted. Its strategy is ecosystem influence pursued at arm's length, complete with the tension that arm's length produces.

Read together, the four describe a market entered from four directions at once — the dedicated platform, the foundational model, the open substrate, the institutional partnership — rather than four contenders for the same prize. The draft Administrative Measures for Digital Virtual Human Information Services, circulated by the Cyberspace Administration in April 2026, tests each wager along a different seam: its prohibition on avatars circumventing facial and voice authentication implicates the payment and identity products at the heart of Alipay, Tencent's financial services, and Baidu's autonomous-driving work; its protection of motion-capture performers speaks directly to the labour history that shadowed A-SOUL; its curbs on virtual intimacy and addictive services for minors constrain the companion products all four are quietly developing; and its call to build out a technology-standards system legitimises the regulatory scaffolding Tencent has helped erect. The combined commercial weight of these firms is real and formative. But the leader's 9.8 percent is the number to keep in view: even the concentrated top of China's digital human industry remains fragmented, and the structure it describes is one in which the dominant vendors set the technological pace and supply the tooling most other deployments depend upon, whilst the greater part of the revenue continues to be fought over, firm by firm, in the long tail below.