China's Digital Human Industry Comes of Age

An industry announces its maturity in the language others use to describe it, and by late 2025 the language around China's digital humans had changed. When People's Daily observed that September that the sector was stepping down from its idol pedestal and deepening its focus on industrial needs, it was marking a transition that the preceding two years had already forced. The animated avatar had begun as spectacle — a virtual singer at a gala, a beauty influencer gathering a million followers in a day — and spectacle is how a novelty is measured. What makes the Chinese sector's coming of age legible is that it stopped being measured that way. Three developments carried it across the threshold: generative AI collapsed the cost of production until the technology became ordinary infrastructure; a brutal consolidation ran directly alongside record expansion, sorting durable firms from hype; and the state, in April 2026, wrote the world's first regulation aimed specifically at digital humans, treating them as consequential enough to govern in their own category. The mark of arrival, in the end, was not a capability but a change of register.

The clearest symptom of an industry still finding its shape is that it cannot yet agree on its own size, and here the disagreement is spectacular. IDC, counting narrowly the revenue of AI-driven digital human platforms and software, put the 2024 market at 41.2 billion yuan. iiMedia Research, drawing a far wider boundary that captures content production, virtual commerce, and downstream services, put the core figure at 339.2 billion yuan for the same year — a difference approaching an order of magnitude between two reputable firms describing the same thing. Any single number offered as the market size is therefore best read against the method behind it, and the disagreement is not a failure of arithmetic but a sign of a field whose edges are still being drawn — where a platform-revenue estimate and a full-ecosystem estimate for the same year can diverge so widely that they scarcely seem to describe the same industry. What is less ambiguous is the structural scale. By the end of 2024 China hosted roughly 1.359 million digital-human-related enterprises, a rise of nearly thirty-seven percent in a year, with more than 413,000 registered in 2024 alone; investment ran to some 110 billion yuan across the year, and patent applications climbed by around 130 percent. The valuation is contested; the underlying activity is not.

The engine of that activity, and the single most important driver of the sector's acceleration, was generative AI. The integration of large language models into digital human systems from 2023 onward compressed production cost and time in a way that changed who could afford to participate. JD.com's platform reduced the cost of a single digital human from tens of thousands of yuan to double digits, a fall exceeding ninety percent; zero-sample techniques introduced in 2024 allowed an avatar to be cloned from as little as thirty seconds of footage and ten seconds of voice; cloud rendering dropped from roughly eighteen yuan a minute to around three; and the entry price for small businesses fell from the half-million-yuan cost of a bespoke project to a few tens of thousands a year, with one 3D ultra-realistic creation tool priced at 199 yuan. Development cycles that once ran to three months shrank to a fortnight. Production speed and price, more than any single visual breakthrough, moved the digital human from studio showpiece to commercial commodity, and they are why the registrations flooded in.

Coming of age, however, was not smooth growth but a culling. Between sixty and seventy percent of the service providers active in 2023 had vanished by the following year, many eliminated, as one industry chief executive put it, because their digital human capabilities could not keep pace with the advance of large language models. Beneath the attrition ran a structural bifurcation that still defines the field: visual realism has sprinted ahead of conversational depth, with insiders rating appearance at seventy to eighty points out of a hundred while scoring interaction quality at only twenty to thirty. The demand-side data counsels caution too, with virtual livestreamers generating less than a fifth of the sales of their human counterparts, average viewing durations falling, and fan attrition running high. And yet the cooling narrative is only half the picture, because the same period produced record enterprise registration, sustained investment, and surging patent activity. Both facts are true at once, and holding them together is precisely what a maturing sector looks like as it separates substance from froth.

The breadth of deployment is the affirmative evidence that the technology had become working equipment rather than demonstration. Finance moved early and deep: by late 2023 eleven banks had fielded virtual digital humans, and one flagship banking avatar was handling more than eighty percent of its institution's call volume. Media matured furthest of all, the synthetic anchor traveling from novelty to routine kit across national and provincial broadcasters. E-commerce became the most visible and most contested arena, where round-the-clock avatar sales streams ran at a tenth of the cost of a human team even as the platforms themselves diverged sharply on tolerance — one major short-video service banning fully AI-driven livestreaming as early as May 2023 and requiring live human oversight. Tourism, heritage, government service, and education followed, from museum guides trained on collections of more than a million artifacts to a set-piece at the 2023 Hangzhou Asian Games, where a giant composite avatar assembled from the digital likenesses of a hundred million participants co-lit the main torch — the first digital torch-lighting in the history of the Games. The deployment mix settled at roughly half media applications, a third service roles, and a fifth industry-specific uses. The avatar had spread across the regulated economy, and it had done so not as a demonstration piece but as ordinary working equipment.

As it spread, it surfaced harms serious enough to draw the state's attention. Around the 2024 Qingming tomb-sweeping festival, a market in the AI resurrection of the dead sprang up, with well over a thousand merchants offering to reanimate the deceased and the families of several late celebrities protesting unauthorized recreations under the civil-code provisions protecting the personality rights of the dead. Fraud syndicates deployed AI face-swapping in romance and investment scams substantial enough to prompt multiple arrests and heavy losses. These were not the growing pains of a curiosity; they were the liabilities of a consequential technology, and they helped pull governance from the general toward the specific.

That governance is the true capstone of the industry's coming of age, because being regulated as a distinct category is itself a form of recognition. The framework accreted in layers over four years — deep synthesis provisions in January 2023, interim measures for generative AI that August, a content-labeling regime in 2025, and draft rules on human-like companion services at the end of that year. The culmination arrived on April 3, 2026, when the Cyberspace Administration published the Administrative Measures for Digital Virtual Human Information Services, a framework of twenty-seven articles across five chapters and the first anywhere written specifically for this technology. It mandates a continuously displayed label, separate consent for using a person's sensitive data, protections for minors against virtual companionship and excessive spending, and graduated penalties, with enforcement shared across more than ten agencies. A parallel scaffold of national standards, including a one-person-one-code identity scheme, hardened the same commitment.

The industry's maturation, then, will not be measured by the aggregate figures that vary by hundreds of billions depending on where the boundary is drawn. It will be measured by whether interactive digital humans can demonstrate sustained engagement, genuine enterprise value, and responsible deployment inside the governance framework China is now assembling. The sector came of age at the moment it became consequential enough to be governed as its own thing rather than as a subset of something larger. What remains to be proven is not that a convincing digital human can be built — that question is settled — but whether the industry can deliver industrial substance in place of the spectacle it has finally outgrown.