Cheng Yixiao, Kuaishou’s chief executive, brought a number to the company’s first-quarter earnings call that investors have been waiting to hear from China’s AI industry: a real revenue figure, growing at a real clip. Kling AI, Kuaishou’s video-generation business, reached an annualized revenue run rate of nearly $500 million in March, up fourfold from a year earlier, Cheng said during the call, according to the company’s transcript.
The figure was the centerpiece of Cheng’s argument that Chinese AI companies can make money. The model has two engines. On one side, businesses pay for API access to generate video at scale — advertising creatives, e-commerce product clips, short-form content for Kuaishou’s own platform. On the other, consumers pay monthly membership fees for generation credits. Cheng described the combination as beginning to prove itself: the commercialization capability of China’s AI application layer is arriving faster than many people expected.
The dual-engine structure matters for how the industry thinks about Chinese AI. Western AI leaders such as OpenAI rely heavily on enterprise subscriptions. China’s app companies are finding that a mix of consumer subscriptions and API sales works in a market where businesses pay for results and individuals pay small amounts monthly. The model suits China’s economics: a massive consumer base, dense mobile payments, and cloud compute priced below Western levels. Video generation, the product being sold, also happens to be the workload where Chinese platforms have distribution advantages — their apps are already full of video.
Cheng’s remarks also touched on cost discipline. Video generation is among the most compute-hungry AI workloads; a single minute of high-resolution generated video consumes far more processing than text tasks. Kuaishou runs its own model training and inference infrastructure, and the company has said it aims to make AI a profit contributor rather than a cost center over time. Executing on that depends on model efficiency improving faster than usage grows, a race the whole industry is running.
The fourfold growth also reflects a wider market shift. Chinese AI companies have moved from giving away services to charging for them, and 2026 has become the year the country’s application layer began producing real revenue in bulk. Subscription tiers at Chinese AI apps are typically priced at tens of yuan a month, and API pricing has been cut repeatedly to win business — a combination that grows volume before it grows margin. Kling sits inside the largest of those bets, because video is the format Chinese platforms know best how to distribute.
Kling AI’s product history gives the number context. Launched in 2024, it was one of the first video-generation models to reach consumers, and it has been updated steadily since. The Kling 3.0 series arrived in February, adding capabilities and reshaping the product line, according to the company. The March run rate — measured a month after the 3.0 launch — suggests the new models are pulling users across the line from free trials to paid plans.
The number also carries a message for Kuaishou’s own business. The company’s core short-video and livestream e-commerce operations generate the bulk of revenue, and AI has been presented to investors as the next growth leg. A fourfold jump in annualized revenue gives that story teeth, and it arrives at a moment when Chinese technology investors are re-rating the sector on AI monetization. Few companies have been able to show the combination Kuaishou is showing: real users, real subscriptions, real API revenue.
Analysts cautioned that a run rate is not a guarantee. Annualized figures extrapolate one month across a year, and subscription churn, promotional pricing and the cost of compute all determine whether the number becomes profit. Video-generation economics are particularly unforgiving — each minute of output carries a heavy compute bill, and price competition among Chinese AI apps is intense. The ARR figure, analysts said, is best read as evidence of demand, not yet evidence of margin.
For the industry, the number is a proof point. It gives the Chinese AI application layer something that has been missing: a consumer-facing AI product converting model capability into revenue at scale, at a pace comparable to the best Western products. The comparisons investors reach for are instructive: Midjourney, one of the most successful paid AI products in the West, has reported annualized revenue in the hundreds of millions of dollars after years of operation, and Kling is approaching that scale roughly two years after launch, in a harder category. Analysts also said the figure understates strategic value: Kling’s models feed Kuaishou’s own content pipeline, giving the company AI capability that rivals would have to buy. The revenue number, in other words, is the visible part of an asset that also improves the core product. For Kuaishou’s shareholders, it answers a question about whether the company’s heavy AI spending is paying off. The answer, on this quarter’s evidence, is that it is beginning to.
