There is not a single wind-shielding and rain-sheltering cloud above Keling's head.

It is reported that Keling is seeking Pre-IPO round financing, with a pre-investment valuation of $18 billion.

According to the financial report, Kuaishou intends to spin off Keling independently to attract external financing. The core reason lies in the two major money - burning areas: computing power and R & D.

In terms of R & D, the expenditure in the first quarter was 3.6 billion yuan, a year - on - year increase of 9.8%. The financial report clearly attributes it to "the increase in employee welfare expenses and the increase in investment in AI".

On the cost of sales side, the bandwidth and server hosting (corresponding to the resource consumption of AI training and inference) cost 1.749 billion yuan, a year - on - year increase of 18.4%; the depreciation of property and equipment and the amortization of intangible assets (corresponding to the depreciation of the previous AI computing power infrastructure included in the current cost) cost 1.775 billion yuan, a year - on - year increase of 43.7%.

Even larger figures are yet to come. Kuaishou's earnings conference revealed that the group's overall capital expenditure will reach about 26 billion yuan in 2026, an increase of about 11 billion yuan compared with last year. The additional part will be prioritized for investments related to Keling and large models.

Despite such large expenditures, most of the revenue that Keling can bring to Kuaishou comes from subscription fees.

As one of the top three domestic video models, this performance is far inferior to Alibaba's HappyHorse and ByteDance's Seedance 2.0. HappyHorse can bring new growth to e - commerce and Alibaba Cloud, while Seedance 2.0 can bring new growth to ByteDance's short dramas and Huoshan Cloud.

Taking Seedance 2.0 as an example, it is reported that the revenue of Huoshan Cloud's MaaS this year has already achieved a ten - fold increase compared with last year's actual revenue. Almost all of this growth is attributed to the explosion of Seedance 2.0.

Obviously, as a member of the first echelon of the top three domestic video models, Keling is currently severely undervalued within Kuaishou's system, which lacks cloud services.

Just as the poor seek fortune far and wide, for Keling to grow, separating from Kuaishou may be the best choice.

Keling Beyond the AI Model

Compared with the visible hard costs, Kuaishou is promoting user growth and paid conversion through a combination of global social communication, KOL creation incentives, and discount promotions.

In other words, the result of each round of Kuaishou's explosive growth depends not only on natural technological drive but also on the platform's unremitting global promotion.

Data from the analysis platform Similarweb shows that since its launch in June 2024, Keling has invested about 2.3 million yuan in three months, mainly for keyword promotion on mainstream platforms such as X and YouTube. The CPC cost of its searches exceeds $0.46.

Multiple phenomena are shifting Keling's marketing focus from search entrances to social promotion.

Firstly, as the video model with the highest quality after Sora, Keling has sparked great enthusiasm on overseas social media. Learning how to register for Keling on one's own and begging for test access have once become hot topics.

Secondly, Elon Musk highly praised Keling's generated content on X, saying that AI is accelerating the entertainment trend. This statement, after being cited by the media, has become a strong endorsement for Keling's overseas expansion.

Thirdly, during the Spring Festival in 2025, at the critical time when Keling's user base was surging and the focus shifted from customer acquisition to customer retention, Keling's "Everything Expansion" template unexpectedly went viral. Users were enthusiastically uploading landmark buildings to generate dynamic videos with a plush toy look.

This makes Keling realize the communication attribute behind the video model. Compared with advertising, infiltrating the short - video space and letting users ask about the production methods themselves may be a more effective way to grow.

This is a path that Sora, Veo, and Runway have never explored. All three of them focus on the enterprise - level market.

For example, Runway has put more than 30 tools in its product, including basic functions like automatic keying and simple cropping, as well as complex ones like green - screen keying, image restoration, and dynamic tracking. Compared with Keling's "one picture, one sentence" approach, Runway hopes to replace AE and PR and become a professional video - editing tool.

As the world's second - largest short - video platform, Kuaishou's genetic advantage means it understands the laws of social communication better. It knows that users are more interested in popular templates that allow them to get started quickly rather than the technical parameters of the video model.

As a result, from pet dancing videos to paper - phone videos, and then to the recent baseball - field videos, Keling has triggered three consecutive global social - media sensations within half a year, each time bringing a sharp increase in downloads and revenue.

From the perspective of traditional exposure - cost logic, Kuaishou has leveraged billions in communication value at zero cost. However, behind this, Keling's investment in creation competitions and creator incentives is soaring.

In 2024, the first Keling Creator Competition had a total prize pool of 300,000 yuan. By the first quarter of 2026, there were nearly ten special competitions for Keling. In the 4K image creation competition, the gold - award prize alone reached $10,000.

In Keling's AI Creator Program 3.0, Keling plans to provide one million yuan in monthly bonuses and ten million inspiration points as incentives and operate it on a regular basis.

Super creators recognized by the platform can receive a black - gold membership worth 1,314 yuan per month, as well as strategic resources such as opportunities to showcase their works at international film festivals.

Reflected in the financial report, Kuaishou's sales and marketing expenses in the first quarter of 2026 reached 10.3 billion yuan, a year - on - year increase of 4.4%. The financial report clearly states that the increase is due to the increase in promotion - activity expenses.

Must It Be Independent Without Cloud Business?

In accounting standards, discount sales should offset revenue rather than be included in sales expenses to avoid the suspicion of inflated revenue.

This rule also applies to Keling, which is in the midst of a price - war pain period.

Keling once tried to raise prices to cope with the rising computing - power costs. For example, the inspiration points required to generate a 5 - second video were 35 in version 1.6. In the Keling 2.0 Master Edition released in April last year, it was raised to 100, but this caused user dissatisfaction, and then it was adjusted back in version 2.1.

In Keling 2.5 Turbo, this price has dropped to 25 inspiration points.

Regarding membership prices, when Keling 2.6 was released, it directly launched a 34% discount on annual membership cards. Since April 1, 2026, Keling has significantly reduced the prices of its AI products. Platinum and higher - level members of the Keling 3.0 model enjoy a 20% discount, and some image functions are even free.

Correspondingly, the pressure from competitors is gradually increasing.

In the domestic market, in April, Alibaba's HappyHorse entered the market with a low - price and chart - topping strategy. Taking 1080P video generation as an example, the membership price of HappyHorse is about $0.78 per second, Keling's is about $0.84 per second, and Seedance 2.0's is about $1 per second. There are reports that ByteDance is expected to launch Seedance 2.0 Lite in the future, with a price of only $0.5 per second.

For Keling, this is a dangerous signal. Compared with Keling, which relies on subscription growth, HappyHorse focuses on how to help the cloud business behind it connect with the Token economy.

In the overseas market, Keling's biggest competitor, Google, also follows a similar cloud - ecosystem logic.

In December 2024, the API price of Google's Veo 2 was still $0.5 per second. In less than a year, Veo 3 dropped to $0.4, and Veo 3 Fast dropped all the way to $0.15, a real halving of the price.

By April this year, Veo 3.1 really brought the price war to the industry's attention - the Fast version for 720p only costs $0.1 per second, and the Lite version only costs $0.05.

This is not just a simple price drop. Google is re - positioning the video model.

Sora's embarrassing exit has served as a wake - up call to the industry - focusing solely on generation quality, pure video - generation applications lack the ticket to the next era. When the product's operating cost far exceeds the revenue and there is no sign of a profit - turning point, exiting is the most rational choice.

Google's approach is to use Veo as an entry point for Google Cloud to acquire customers, open it to paid developers through Google AI Studio, and replace membership subscriptions with token - call billing. As long as the sales of Google Cloud's computing power and storage increase, the profit situation of Veo itself can be ignored.

In other words, Google doesn't want to sell video - generation capabilities but rather to offer a video product on Google Cloud and sell it to enterprises and developers.

Keling's business model is to build a more open ecosystem and select professional creators from it. Kuaishou CEO Cheng Yixiao once said in an earnings conference that nearly 70% of Keling's revenue comes from P - end subscription members, that is, professional creators between ordinary consumers and enterprises.

This is why Keling is so keen on large - scale social - media breakthroughs. When its competitors are fighting on both cloud and model fronts, Keling has no way back and must go all the way.

Caught in a Dilemma

"The real opportunities in this track may only belong to a few early - movers who can defend their moats," an AI practitioner told "Market Insights".

This is also the challenge that Keling faces.

According to "Intelligent Emergence", Seedance 2.0's monthly revenue has exceeded 1 billion yuan. In contrast, Keling's revenue in the entire first quarter of this year exceeded 650 million yuan.

"All products are released while being iterated," a industry observer said. Enterprises need to continuously prove their strength to the capital market and users with new models, and user expectations are also changing, from focusing on top - tier products to using the cheapest ones.

Olivia Moore, a partner at venture - capital firm a16z, once analyzed the reasons for Sora's failure - the first - day user retention rate was 10%, the 30 - day user retention rate was only 1%, and the 60 - day retention rate dropped to 0:

"The model did have viral spread, but most users did not continue to use the application."

This fits the impression of most users of video models - they are excited about new features for a day or two and then quickly disappear.

Data from Sequoia Capital shows that the first - month retention rate of traditional applications is about 63%, while that of AI applications is only 42%, and the daily activity is even lower.

Kuaishou's solution is to continuously create the next popular content and use content maintenance to avoid the freshness trap.

In an official statement, Ma Hongbin, Kuaishou's senior vice - president, defined Keling as a new form of equal - access productivity: "The ultimate value of AI is to return the right of expression to everyone." The ultimate goal is to break through technological barriers and achieve equal - access creation.

Maintaining user retention requires continuous investment.

The first aspect is the computing - power cost. Keling's annual revenue in 2025 was 1.04 billion yuan, while Kuaishou is expected to invest an additional 11 billion yuan in large models such as Keling in 2026. Using tens of billions of capital investment to support a product with a revenue in the billions is what Kuaishou is currently doing.

The second aspect is creator incentives. Behind nearly ten special competitions in a quarter, Kuaishou urgently needs to increase Keling's exposure at the P - end. It not only provides money and inspiration points but also long - term strategic cooperation resources.

This is essentially an investment in user assets. By investing in professional creators, generating content with the Keling label, and further spreading it at the C - end. When ByteDance's Jiemeng and Alibaba's HappyHorse enter the market to compete for creators, Keling faces the pressure of continuous investment. It either increases the investment or watches the creators leave.

The third aspect is price subsidies. The $0.05 - per - second Veo 3.1 Lite is not even Google's full - force offering. In extreme cases, Google can allow video - inference losses as long as it meets customers' computing - power and storage needs, and the overall account will still be profitable.

Without the support of a cloud empire, the more the industry breaks the price perception of AI videos, the more it challenges the positioning of independent video models.

This forces Keling to further strengthen its base of P - end creators, enhance brand awareness through more social communication, and let professional creators deposit more material designs on the platform to increase the migration cost, thus building an ecological moat for the independent product.

This will affect the market's judgment on Keling's independence in the next step. Whether a spun - off Keling can continue to grow depends on how many users' usage habits it can accumulate, rather than how low it can reduce the generation price.

This article is from the WeChat official account "Market Insights", author: Jing Xing. Republished by 36Kr with permission.