On October 16, 2026, Alibaba Cloud's official website quietly posted a product change announcement: virtual digital humans had officially begun the delisting process. New purchases would cease on November 30, renewal channels would close in May 2027, and by November 2027, this product—once hailed as metaverse infrastructure—would be fully discontinued, with the console, ticket channels, and supporting technical services all taken offline. No dedicated launch event, no industry farewell forum, no media press release—just a plain line of announcement text, bringing an official close to a raucous industry cycle.

Cast your mind back exactly five years, to the late autumn of 2021, when a virtual beauty blogger named Liu Yexi posted a two-minute-plus short video on Douyin. Overnight, she gained over a million followers; the comment section was packed with exclamations, and brands, investors, and startup teams swarmed in, chasing the traffic. Later, her team reviewed this sudden explosion and offered a weighty breakdown: fifty percent of the dividend came from the metaverse hype, thirty percent relied on digital human rendering and motion capture technical capabilities, and the remaining twenty percent was the creative narrative of the short film itself. This review also planted, early on, the fatal foreshadowing of the entire track: the rise and fall of virtual digital humans was never simply about technical capability—their fate was always standing on the hype.

On one side, the big companies were delisting in an orderly fashion, liquidating their businesses, packaging and archiving the "next-generation interaction carrier" they once held high hopes for; on the other side, five years earlier, the whole internet was in carnival, BP decks flying everywhere, everyone convinced digital humans would become a brand standard, a new traffic entrance, the ultimate replacement for real people. In this enormous temporal contrast lie the three successive deaths that occurred on this track.

It was not a sudden collapse, but a slow death, layer by layer, along three main lines: technological iteration, capital narrative, and commercial implementation.

The first layer of value of virtual digital humans was the technological spectacle.

Why could Liu Yexi become popular? Because during that period, "looking like a real person" was itself a scarce capability. Hyper-realistic modeling required top-tier CG skills; motion capture equipment cost millions; and facial rigging, hair solving, subsurface scattering—each was hard craft. In 2021, a virtual face that could withstand a close-up was backed by a long production pipeline. Audiences were willing to stay because they were witnessing a spectacle: a nearly indistinguishable person, living on the screen.

Technological scarcity propped up the entire industry's valuation. Brands scrambled to claim virtual spokespersons—beauty, banking, FMCG, every one had to launch a flawless face. AYAYI debuted on Xiaohongshu, collaborated with Guerlain, and became the first person of the metaverse; hyper-realistic virtual humans were released in dense waves, every face emphasizing the same thing: we are more like real people than real people.

But spectacle has a natural shelf life—it had no real industrial foundation or audience, and once the novelty passed, it lost its appeal. Virtual humans at this stage died from two forces.

The first was aesthetic fatigue. The greatest shortcoming of virtual humans was hidden beneath the glamorous exterior from the day they were born: exquisite shell, hollow soul. Most could only read pre-written scripts, with restrained and stiff expressions, eyes lacking the warmth of flow—so-called interaction was merely jumps between preset phrases. Audience amazement came from looking like real people, but after they had seen dozens of similar, standardized beautiful faces, wonder quickly turned to numbness. Real streamers get tired, say the wrong things, have improvised moments of emotion—they feel more real. In virtual livestream rooms, average viewer stay time slid all the way down, and platforms slowly withdrew the tilted traffic.

The second was the democratization of technology itself. In 2023, AIGC exploded. Before, hyper-realism was a craft requiring years of accumulation; after, generative tools turned "looking like a real person" into something you could buy for a few bucks. Once the threshold was flattened, the technology premium approached zero.

Many CG company founders complained that the company's core asset—the hyper-realistic production pipeline—was being replaced by AI tools far faster than expected. Previously clients paid for realism; now they generate it themselves with software.

Technological iteration killed the myth of technology. When looking like a real person became a free and egalitarian capability, the first layer of value of virtual humans returned to zero. This was the first death of digital humans. No farewell announcement, no discontinuation statement—brand virtual spokespersons simply faded from homepages, no longer updated, no longer appearing.

The second death happened on the balance sheet.

At the height of the boom, the industry had two parallel discourses. One faced brands and investors: virtual humans never have scandals, broadcast around the clock, replicate infinitely, the ultimate low-cost marketing solution. The other was hidden in studio bills: a premium short film cost hundreds of thousands; a hyper-realistic image from modeling, rigging to motion capture and rendering cost millions; subsequent scripts, operations, middlemen, and content iteration were endless ongoing expenses.

Capital bought in initially because it believed a bigger story: virtual humans were the entrance to the metaverse. During the hottest metaverse narrative from 2021 to 2022, virtual digital humans were defined as the aboriginals, spokespersons, the first residents of that virtual world. The story was big enough to temporarily ignore the commercial loop. Financing round after round, valuations rising with the tide, everyone defaulting to the same logic: burn money to occupy the position first, the future will always pay off.

But this story had a fatal flaw: it could not monetize.

High-end custom virtual idols had few clients, high unit prices but long delivery cycles, hard to scale; advertising endorsements, livestream commerce, virtual concerts—every monetization path that could be counted was tried, none could cover costs. One founder who raised financing said their virtual idol took on a dozen brand collaborations a year, which sounded lively, but the labor cost of one offline concert equaled three quarters of endorsement income. Virtual humans were an expensive, operations-heavy content business, and the vast majority of players simply did not have the ability to operate content.

In 2023, two pillars broke simultaneously. One was the capital turn. ChatGPT ignited the large model frenzy, and investors found a sexier next-generation entrance: no modeling, no motion capture, no middlemen needed—the large model itself was interaction. Virtual digital humans were downgraded from metaverse entrance to large model appendage, even mocked as old-era PPT. After that, financing heat on the virtual human track plummeted off a cliff. One industry insider said that after Series B, he met with over sixty institutions, but no one was willing to listen to stories anymore.

The other was big company contraction. Pico layoffs, XR business adjustments, Tencent and Alibaba successively withdrawing from or shutting down virtual digital human-related business lines. Even the companies least short of money were unwilling to pay for a business without clear returns, and many startups disbanded and sought other ways out.

The capital tide receded, the bubble lost its support. Those startups surviving on financing fell with it: R&D teams cut, motion capture studios idle, projects originally planned to mass-produce virtual idols shelved. What died this time was the commercial fantasy of low-cost star-making and rapid payback.

Capital never bought technology, but the narrative of the next-generation entrance. When the narrative went bankrupt, the companies went bankrupt with it.

After the first two deaths, the track should have been cleaned up, but speculators took the last baton.

When the refined idol route failed, when big companies exited and financing was hopeless, the survivors turned to a cruder path: unattended digital human livestreaming. Merchants bought template images, imported looping scripts, ran them 24/7, no one needed to watch, no streamer salary to pay, fantasizing about picking up traffic through endless online hours. For a time, platforms were flooded with digital human livestream rooms with repetitive movements, mechanical tones, and the same segment looping back and forth.

This was the most speculative stage of the entire industry. Many entered simply to find a way to sell goods while lying flat without investing manpower. The image on screen repeated selling points over and over, the scene looping, no real responses, no live communication—essentially a looping recording dressed in the cloak of real-time livestreaming. A price war followed: hyper-realistic customization fell from millions to hundreds of thousands, 2D templates from tens of thousands to hundreds, dozens, even monthly low-price subscriptions, profits compressed to near disappearance.

The degradation of the business model ultimately backfired on the entire industry.

On one hand, credibility collapsed. When virtual humans in public discourse gradually became synonymous with electronic puppets and leek-cutting, companies seriously doing technology were caught in the blast. On the other hand, regulation and platform rules tightened. Labeling regulations for deep synthesis content landed, the Interim Measures for the Management of Generative AI Services took effect, platforms concentrated on cleaning up looping recordings and low-quality unattended livestream rooms, traffic limits, weight reductions, and account bans followed. So the cheapest, easiest path was blocked. Many small and medium merchants exited completely at this point—they spent a few thousand bucks on a system, expecting nonstop orders day and night, and ended up with few views, extremely low conversion, even account penalties. So-called zero-manpower livestreaming proved to be a bet, and what lost was the entire industry's reputation.

In short, speculators squeezed out the last value of virtual humans at the lowest cost, and personally buried the industry's last credibility. This was the third death. After the noise faded, what remained were only a few technical applications rooted in specific scenarios: customer service guides, government affairs broadcasts, corporate showrooms. They were quiet, restrained, no longer using replacing real people or igniting traffic as slogans, just silently doing the job of a tool.

But dialectically, the virtual digital human technology did not die. In fact, it came back in another way: in the large model era, everyone can generate their own digital avatar, AI anchors and digital employees re-entering corporate procurement lists. This time, no one promises never having scandals or never getting tired—it just serves as a tool, quietly staying where it truly fits.

What truly died was that once-explosive industrial myth that promised wealth creation. It died from the shelf life of technological spectacle, from the bankruptcy of the capital narrative, and from the dusk of speculators.

The three deaths buried in turn the technology premium, capital patience, and industry credibility. Years later, when a new hype rises again, we may remember this lesson: the destruction of an industry often comes because too many people are eager to extract value from it.