In the summer of 2024, there was only one way to introduce a Chinese video model to the rest of the world, and it was by reference to an American one. Kling was the rival from across the Pacific, the thing you could already try while the other one was still a waitlist, and the reference is what made it famous. Two years on, the reference has come around the circle. OpenAI has announced it is shutting down Sora, and Kuaishou has picked banks for a Hong Kong listing that Bloomberg puts at at least a billion dollars. The company that once needed an American name to be legible is now being priced on its own books, and the flip is the whole story.
Start with the company, because the company is the under-appreciated part of the story. Kuaishou is China's second-largest short-video platform, and it is the one the Chinese themselves use more, a fact that matters, because the app that a country's own creators live in is the distribution channel no one else can buy. The name is a transliteration of the Chinese for "fast hand," and the product was built around exactly that, the low-tier city user, the maker who films with one take and posts before the moment passes. That is a strange foundation for a frontier model company, and it is the right one, because the people who live in the feed know what a clip needs to survive the feed. The model was not built to impress a benchmark; it was built to be usable by the person holding the phone, and the person holding the phone is the one the API eventually serves through somebody else's product.
Now the fireplace. Jason Zada, the director behind Coca-Cola's 2024 AI holiday ad, used Kling for a YouTube bit in which a fire burns quietly while Santa, turkeys, and astronauts make their inexplicable appearances. The joke works because the fireplace stays familiar; you can leave it running in the background and look up just in time to wonder what the astronaut is doing there. A fireplace is forgiving. A multi-shot scene is not. If a face changes when the camera cuts closer, the viewer notices, and if the background shifts every time the editor asks for another angle, the separately convincing clips never fit together. That gap is what Kling 3.0 closed in February, with control across shots and characters that carry through them. A model that makes a good clip gives you something to post. A model that gives you the shot you need gives you a reason to come back to the edit, and the coming back is where the money is.
The numbers, in the order they matter
The first number is the one that came before the shutdown, and it explains everything after. In December 2025, while Sora was still running, Kling brought in more than twenty million dollars in a single month. The second number is the one that came after it: when the shutdown landed, global weekly active users on the mobile app rose about four percent on the week, per Sensor Tower data that Bloomberg cited. Four percent sounds modest, and it is, by the standards of the week, which is the point. The category was never about the people who try a video model; it is about the people who are mid-job and need the next shot, and the four percent is the fraction of them who crossed over in seven days. The third number is the quarterly one. Kuaishou's interim report put Kling's second-quarter revenue above 850 million yuan, more than triple the level a year earlier. And the fourth is the structure: the summer restructuring moved the business into a standalone subsidiary, the outside money came in, and by the end of October the round had reached nearly three billion dollars, which is the number the listing is the exit of.
The same day, the field moves
The listing lands in the middle of a week that is not gentle on anyone. ByteDance opened Seedance 2.5's 4K video generation API to developers on the same day, which is to say, the spec war did not wait for the pricing to settle, and it will not wait. Google's Veo is running the American side of the race, and Runway open-sourced its Praxis-1 this week, which is the third move in the same category, the one where the tooling is going under the model and the model is going into the tool. The listing does not need any of that to be right. It needs the editor at eleven at night to keep opening the same app, and that habit is the one the 850 million yuan is the quarterly print of. What the competition changes is the slope, and the slope is the one the public price will argue about after the first day of trading, not before it.
What it means
The reference flipped, and the flip is a milestone, because the reference is the moat. A company that has to be introduced by reference to a competitor is still living in the competitor's category. The next video model from the next lab will be introduced as the rival to Kling, not the rival to Sora, and a rival to a company with a listing is a harder reference to be. A listing is not a thing you can shut down, and the category organizes itself around the one that cannot be.
The distribution is the part of the business the user never sees. The API embeds Kling's generation inside other companies' products, and the person who clicks generate sees the other company's name on the button. The other company's name is the one that gets remembered, and the one that gets remembered is the one that ships. Kuaishou built this on top of years of labelling video and figuring out who wants to watch what, and the result travels further than the feed: an American director can use the model with no reason to watch the app, and the app's own numbers are the receipts for it.
The IPO is the exit the gray market does not have. The private round set the number at three billion; the listing is where the public market tests it, and the public price is the one that does not move. Every round the competitors raise, every spec they ship, prices against that number. The spec war between Google, ByteDance, and the rest is still running, and another model release can change what looks impressive by the end of the week. What takes longer to change is how a person works when the job is already underway. And the venue deserves a line of its own, because a Chinese frontier AI company pricing itself in Hong Kong, in the same week the yuan is overtaking the dollar in the city's bond listings, is not a coincidence of geography. It is the hub doing the job it has been rebuilding for three years, which is being the place where Chinese technology capital gets a public number, and the number is the one the next round, the next listing, and the next spec are measured against.
The American side of the story is the absence. The Western press is writing about the shutdown; the Chinese company is the subject of the other story, the one that is being priced. The shutdown is the background, and the background is what the listing is measured against. An absence can be a reference point, but only for a while. The reference now points the other way, and it is the company with the banks that holds the end of the rope.
🔥 Hot Takes
1. Sora's shutdown is the best press release Kling ever got, and OpenAI did not charge for it. The shutdown is not a category loss, it is a transfer, and the transfer is the four percent. The fraction that crossed over in a week now has the first shot on the timeline, and the person with the first shot on the timeline is the one opening Kling in the morning. OpenAI bought Kling a distribution event at zero cost, and the invoice is the 850 million yuan that is showing up in the quarterly print.
2. The invisible brand is the Trojan horse, and the horse is the API. The user who clicks generate never sees the Kuaishou name, which means the distribution is already inside products the user will not leave. Demos get screens; embedding gets the timeline. The company that owns the timeline owns the workflow, and the workflow is the moat that no spec release, however impressive, can reach in time.
3. The IPO is the moment the public market gets its price on a Chinese video model, and that price changes the whole field. A three-billion private round is a claim. A listing is a number that does not move, and every competitor's next round, next spec, and next demo is measured against it. The Seedance 4K API opening this week is the spec war continuing; the listing is the thing that ends the argument about what the category is worth, because after the first day of trading, it is not an argument anymore, it is a quote.
So the reference flipped, and the flip is the milestone. The company that was introduced as the rival to a model that is now shut down is the one with the banks, the round, and the listing, and the listing is the number the rest of the field is measured against. The editor at eleven at night is not choosing between demos. The editor is opening the tool that does the second shot, and the tool that does the second shot is the one that was priced this week. The dead model is the reference now. The living one is the price.