---
title: "Kuaishou's 'AI Commercial Internal Cycle' is now operational."
type: "Topics"
locale: "zh-HK"
url: "https://longbridge.com/zh-HK/topics/33240516.md"
description: "Whenever groundbreaking new technologies are mentioned, most people's minds conjure up images of overwhelming disruption: the sound of the steam engine quickly toppling agricultural civilization with the Industrial Revolution; the birth of the internet sweeping the globe and overturning traditional lifestyles. But in reality, truly disruptive technologies that take root and change the world rarely achieve success overnight through pure 'replacement.' More often, they coexist and grow alongside traditional models in the process of improving efficiency and reducing costs, ultimately reshaping the landscape quietly. The mid-year reports of Google and Meta this year are the best examples..."
datetime: "2025-08-22T08:42:20.000Z"
locales:
  - [en](https://longbridge.com/en/topics/33240516.md)
  - [zh-CN](https://longbridge.com/zh-CN/topics/33240516.md)
  - [zh-HK](https://longbridge.com/zh-HK/topics/33240516.md)
author: "[锦缎研究院](https://longbridge.com/zh-HK/profiles/2576456.md)"
---

> 支持的語言: [English](https://longbridge.com/en/topics/33240516.md) | [简体中文](https://longbridge.com/zh-CN/topics/33240516.md)


# Kuaishou's 'AI Commercial Internal Cycle' is now operational.

Whenever epoch-making new technologies are mentioned, most people envision a scene of overwhelming disruption: the steam engine roars, and the industrial revolution quickly topples agricultural civilization; the birth of the internet sweeps the globe, overturning traditional lifestyles.

In reality, truly disruptive technologies that can take root, grow, and change the world rarely achieve this through pure "replacement" in one fell swoop. They often coexist and grow with traditional models by improving efficiency and reducing costs, ultimately quietly reshaping the landscape.

Google and Meta during this year's interim report period are the best examples. Even with limited commercialization of pure AI products, they both significantly beat market expectations, relying on the rapid advancement of AI-enhanced advertising businesses like Advantage+, Reel, and Pmax.

Focusing on the domestic market, the current main logic of AI development and application is the same. Just yesterday, Kuaishou, a leading Chinese video model, released its 2025 interim report: operating revenue recorded 35.046 billion, a year-on-year increase of 13.1%, and adjusted profit was 5.618 billion, a year-on-year increase of 21.6%, with both revenue and profit significantly beating expectations.

Moreover, in terms of future expectations, Kuaishou also shows relatively optimistic signals for AI development and overall business growth. CFO Jin Bing revealed at the performance meeting that Keling AI's full-year revenue in 2025 is expected to double from the target at the beginning of the year, which strengthens Kuaishou's determination to make long-term investments in Keling AI. It is expected that the investment in Capex related to Keling AI in 2025 will also double compared to the initial budget.

At the same time, Kuaishou announced the distribution of its first special dividend since listing, 0.46 Hong Kong dollars per share, totaling about 2 billion Hong Kong dollars, while rewarding shareholders, it also shows the capital market its confidence in the new cycle. After the financial report performance meeting, the capital market also gave positive feedback to Kuaishou's better-than-expected interim report, with Kuaishou-W (HK1024) rising nearly 5% during today's trading session.

Kuaishou's excellent interim report provides us with an advanced sample of how AI deeply transforms a mature business ecosystem: based on this financial report, we will explore in detail how AI reshapes the existing short video business ecosystem from both supply and demand sides, as well as from the two dimensions of content and e-commerce, in a "2X2" four aspects.

## **01 Demand Breakthrough: AI Opens Kuaishou's Growth Ceiling**

According to the China Internet Development Status Survey Report, from the second half of last year to the beginning of this year, due to the growth ceiling of internet devices, the overall user usage rate of online videos showed a high-level decline over the past 18 months, dropping from 97.7% to 96.7%.

 Figure: The scale and usage rate of online video users in the past two years, source: China Internet Network Information Center

Despite the overall industry trend still being weak, Kuaishou presents a "unique" picture, as shown in the interim report:

**In the second quarter of this year, Kuaishou's DAU and MAU reached 409 million and 715 million, respectively, with year-on-year growth of 3.4% and 3.3%, respectively. The average DAU hit a new high and exceeded 400 million for four consecutive quarters. The average usage time reached 126.8 minutes, and the total user usage time increased by 7.5% year-on-year.**

The steady growth of Kuaishou's core operating data is fundamentally due to the improvement of the recommendation mechanism through AI technology, achieving a better match between content supply and user demand:

In a stable user base, the relationship between content supply expansion and demand growth is not linear, especially for short video content platforms, which are essentially a non-symmetrical supply-demand business model. For example, users who like to watch funny videos will not increase their demand even if there is more popular science content supply.

Indeed, similar to the supply of physical goods, the number of content supplies is not the core obstacle to demand. The core lies in whether the optimization of supply-demand matching can improve user stickiness, thereby driving the growth of overall usage time and increasing total traffic.

Kuaishou has found an effective way to improve supply-demand matching on the content side through the application of AI technology, such as the OneRec end-to-end recommendation model that has been applied on both ends of Kuaishou (APP, Speed Edition).

Traditional recommendation models are generally a "multi-stage pipeline" architecture, with the core problem being that each layer of the model only manages its own part, making it difficult to consider the user's real interests from an overall perspective. Moreover, multi-layer models create constraints, needing to increase click-through rates while also considering niche and diversity, ultimately leading to continuous deterioration of consistency and efficiency.

OneRec essentially switches the recommendation problem from multi-stage, layered selection to end-to-end large model generation. In the past, recommendation systems were like assembly lines, filtering content layer by layer; now, OneRec directly generates a list of video content that users are most likely to be interested in through a large model.

**Overall, the OneRec end-to-end large model solves two core problems:**

First is cost and efficiency.

On the cost side, through architectural innovation, the utilization rate of computing power for training and inference (MFU) soared to 23.7% and 28.8%, respectively, and the operating cost (OPEX) was only 10.6% of the traditional solution. On the efficiency side, the computing power utilization rate of large language models on top chips like H100 can generally reach over 40%.

Second is recommendation effectiveness.

OneRec introduces a multimodal semantic tokenizer in content understanding. It not only looks at video tags, titles, or simple visual features but breaks down each video into more detailed information fragments—visual content, text, voiceover, music, and even user interactions.

For example, a food short video, the tokenizer will not only capture ingredients, cooking actions, and kitchen environment in the picture but also analyze the conversational tone in the voiceover, user comments, and bullet screens, all becoming part of content understanding. After a more detailed and three-dimensional understanding of user behavior, it naturally clarifies user preferences, thereby improving the logic of inference.

We can directly see the effect. Kuaishou conducted a 5% traffic AB test for a week after OneRec went live, similar to head-to-head experiments in the pharmaceutical industry, and achieved positive results in all interaction indicators such as likes, follows, and comments.

OneRec also performed impressively in Kuaishou's local life service scenarios: AB comparison experiments showed that the solution boosted GMV by 21.01%, order volume by 17.89%, and the number of purchasing users by 18.58%, with new customer acquisition efficiency achieving a significant increase of 23.02%.

**Users being able to see content and users being able to see the content they want to see, the difference of two words, a world of difference. Naturally, it can help Kuaishou enhance user stickiness and increase total traffic.**

Let's talk about the product demand side, that is, how Kuaishou is gradually reconstructing the decision-making chain of content e-commerce through AI technology.

The core of maintaining the commercial chain of traditional shelf e-commerce is search, that is, people looking for goods; the core of maintaining the commercial chain of content e-commerce is the recommendation algorithm, that is, goods looking for people. Relying on the capabilities of large models, Kuaishou not only achieves more precise goods looking for people but also realizes AI-based services and improves the consumption chain on the basis of further perfecting and evolving the recommendation algorithm.

For example, buying a piece of clothing, using the AI fitting function based on AIGC technology on Kuaishou's APP can more intuitively show users the effect of wearing it, shortening the traditional steps of planting grass - placing an order - receiving goods - trying on - returning and exchanging goods. I personally tested it, and the effect is very good.  

Figure: Kuaishou APP fitting function illustration, source: Kuaishou APP 

Thanks to the optimization of AI empowerment on the product demand side, the consumption decision chain is significantly shortened, and the number of active buyers in Kuaishou e-commerce in the second quarter reached 134 million, and the repurchase frequency of active e-commerce users is also continuously increasing.

At the same time, Kuaishou also applies AI technology to the matching of people and goods and the pre-sale and after-sale experience. Based on the capabilities of large models, Kuaishou can accurately predict user purchasing behavior and preferences, automatically identify strongly related categories and interest tags, and make the algorithm better understand users' potential needs.

Kuaishou's intelligent customer service for pre-sale and after-sale services is also far more intelligent than traditional automatic customer service, capable of solving many high-frequency problems faced by merchants daily. For example, in this year's 618 case of Cui Green Jewelry, relying on Kuaishou AI Assistant, a large number of high-frequency, professional product information was answered for users before and after sales, solving 60%-70% of routine problems, greatly reducing the pressure on manual customer service.

**The amplification of the product side through AI technology is also directly reflected in the financial report, with Kuaishou's overall e-commerce GMV recording 358.9 billion in the second quarter, a year-on-year increase of 17.6%.**

In summary, one of the core reasons why Kuaishou can still achieve impressive growth during the bottleneck period of content platforms is that it has achieved a virtuous cycle on the content level through AI technology and achieved precise reach of user product demand, thereby breaking through the expected ceiling on the demand side.

## **02 Supply Innovation: AI Reshapes the Short Video Business Ecosystem**

Following the logic of our previous article, if viewed from a business perspective, the supply side is also divided into product supply and content supply. The most direct account manifestation of supply optimization is that **Kuaishou disclosed during the interim report period that the number of new merchants entering Kuaishou in the second quarter increased by 50% year-on-year, and the daily average number of active merchants in the mall increased by 30% year-on-year.**

In the current background of platform differentiation and division, Kuaishou can still capture incremental supply on the supply side. In addition to the fact that Kuaishou is currently one of the few platforms that can increase total traffic, more importantly, Kuaishou continuously alleviates the two core pain points of e-commerce merchants through technology empowerment: high marketing costs and operating costs.

**The first difficulty is that the cost of marketing materials cannot be effectively controlled. E-commerce marketing is fast-paced and costly. If marketing materials cannot keep up with traffic trends, the effect of placement will be unsatisfactory.**

However, the emergence of AI technology applied by Kuaishou solves this problem. For example, the Magnetic Creation Platform mentioned at this year's Magnetic Conference allows merchants to edit popular video counterparts with one click, intelligently generate high-quality scripts, and create digital human videos, obtaining higher-quality placement materials at lower costs and faster speeds, effectively solving the contradiction between cost and efficiency.

Another example is the Nuwa digital human, which supports virtual humans broadcasting 24 hours a day and has matrix operation capabilities, truly achieving "low threshold, high efficiency" in live marketing.

According to the annual report season, Kuaishou CEO revealed in a conference call that Kuaishou's internal calculations estimate that AI large models can reduce the production cost of customers' short video marketing materials by 60-70% or even more.

**The second difficulty is that the conversion rate of promotional traffic is generally not high, and it is difficult for promotional ads and natural traffic to interact and match.**

Kuaishou relies on AI capabilities to launch the UAX (Universal Auto X, fully automatic placement) solution, which has four core capabilities: intelligent infrastructure, intelligent control, intelligent creativity, and intelligent targeting, effectively assisting customers and agents in enhancing volume capabilities, stabilizing volume costs, and improving placement efficiency.

According to Jiang Peng, head of Kuaishou's commercialization algorithm, the decision-making learning of the marketing recommendation large model's chain of thought can allow the model to gradually decompose and summarize complex problems like humans, infer users' purchasing intentions and product marketing points, and more precise matching of people and goods leads to a 10% increase in product order volume and a 20% increase in new product consumption.

The UAX fully automatic placement solution completed the upgrade from rule-based decision-making to model-based decision-making. In the second quarter of this year, the proportion of marketing consumption of UAX in the total consumption of the external cycle increased to about 65%.

**Optimizing ROI and solving the blockage of placement conversion rates naturally increases merchants' inclination to place ads and also drives budget tilts. Kuaishou's overall online marketing revenue recorded 19.8 billion in the first half of the year, a year-on-year increase of 12.8%, which is the best proof and result.**

**The third difficulty is that the traditional chain from placement to business landing is very long, and merchants need a lot of infrastructure work to maintain the business conversion brought by placement. For some small and medium-sized merchants, it is difficult to have enough business experience and labor costs to match the actual business of placement conversion.**

Kuaishou solves this problem from both technical and policy dimensions. On the one hand, the low-cost AI tools mentioned in the previous article are provided to merchants from material production, advertising placement, to live broadcast product introduction, and intelligent customer service, helping merchants undertake the business growth driven by placement.

On the other hand, Kuaishou has provided real money subsidies. For novice merchants, Kuaishou e-commerce has launched a new merchant cold start plan, focusing on targeted support and growth for new merchants within 90 days of their first broadcast.

The owner of the shoe and boot anchor Fan Xiaoluo started operating Kuaishou at the end of February this year. As of May, the account's cumulative GMV has exceeded 10 million. At the beginning of Fan Xiaoluo's entry, she ensured a daily live broadcast duration of 5-7 hours, achieving growth with the exclusive support of the first broadcast for 30 days and the continuous support of high-quality new merchants for 90 days.

At the same time, by utilizing Kuaishou's cross-domain traffic resources, through explosive slice and short video planting, cross-domain traffic intercommunication is achieved, with more than 30% of transactions coming from short video diversion. The full-site promotion also helped Fan Xiaoluo achieve faster growth, keeping the placement ROI stable at around 30.

In summary, Kuaishou uses AI technology to solve the three major problems that traditional merchants may face, naturally leading to a prosperous ecosystem on the supply side, combined with the dual effects of optimized matching on the demand side, ultimately promoting the rapid growth of GMV.

## **03 Ace "Keling": The High-Level Narrative of Kuaishou's Value Transformation**

As we write, we have roughly clarified the logic of AI changing supply, demand, content, and products revealed during Kuaishou's interim report period, and it is not difficult to understand that Kuaishou's growth is actually no different from the logic shown in Meta and Google's second-quarter financial reports.

**However, Kuaishou's AI narrative goes far beyond this. In the context of the global AI competition narrative, its ace AI application—"Keling AI"—has a value that is actually far higher than the general market perception, which is also an important variable for Kuaishou's further value transformation.**

First, Keling AI is a key force for China to participate in the global AI application peak competition. China needs iconic AI applications to benchmark against global AI frontier applications, and Keling AI's capabilities are outstanding.

In the current multimodal field of AI applications, there are strong competitors, with domestic players like Keling AI, Jimeng, and Hailuo, and overseas players like the established multimodal products Runway, Sora, and Stability AI.

All technological applications start with a small group of technology enthusiasts and early adopters becoming opinion leaders, but as the industry continues to develop, more professionals can explore truly valuable technological innovations from more angles and details.

At this year's WAIC, Keling AI's capabilities were widely recognized by professionals. Abela, CEO of the globally renowned creative material platform Freepik, said, "We have integrated more than 10 different video generation large models, and the number of videos generated by users using Keling AI exceeds the total of other models combined."

Clearly, professionals have strongly confirmed Keling AI's technological leadership and user recognition on a global scale, establishing its representative status as a Chinese AI application.

Secondly, Keling AI is continuously proving its enormous value as advanced productivity to the market.

The vitality of AI technology lies in whether it can truly transform into an engine that drives industrial upgrading and social progress. We can see that Keling AI is vividly illustrating this through extensive commercial landing scenarios, such as the production of advertising materials mentioned earlier, and the scenario of industrialization landing in the film industry:

On June 25, the world's first AI unit story collection "New World Loading" held its premiere in Beijing. The work was co-produced by Kuaishou Keling AI and Xingmang Short Drama, and produced by the Outliers team, with a total duration of 180 minutes. The drama is also currently known as the AIGC drama work with the widest range of subjects and the largest narrative volume, with a global cumulative view count of 200 million.

Even in the game industry, there are many landing scenarios for Keling AI. The popular mobile game "Nishuihan" has reached a deep cooperation with Keling AI, embedding the industry's leading AI video generation capabilities into the game, enriching the game's social gameplay.

Data shows that as of July this year, Keling AI has produced more than 200 million videos and 400 million images, serving over 20,000 corporate clients. Cheng Yixiao also mentioned in a conference call, "In the future, we hope that with the improvement of the comprehensive performance of the model, Keling AI will gradually have the ability to deeply apply in industrial-grade scenarios such as game production and professional film production, expanding more industry users."

Finally, Keling AI, through its good commercialization performance, not only enhances its valuation but also shapes a virtuous cycle model of capital expenditure for Chinese AI applications.

In the second quarter of this year, Keling AI's operating income exceeded 250 million yuan. Nowadays, few AI applications can confidently disclose the progress of commercialization income. Keling AI confidently delivers a commercial answer sheet, not only proving its actual product strength but also demonstrating to the market the potential and value of AI application commercialization.

According to CICC research report data, the "Keling AI" series of models have a market share of 30.7% in the global AI video tool market, firmly ranking first in the industry. According to CFO Jin Bing at the performance meeting, Keling AI's full-year revenue in 2025 is expected to double from the target at the beginning of the year.

Relying on the excellent commercialization performance of Keling AI, Kuaishou has also set a precedent for upgrading capital expenditure in the Chinese AI field. CFO Jin Bing revealed that **the Capex investment related to Keling AI has also doubled compared to the initial budget, and considering that the gross profit margin at the inference computing power level of Keling AI has been achieved and remains stable, the additional computing power investment has a controllable impact on the group's overall profit, and capital expenditure has entered a virtuous cycle.**

We have analyzed many times that the key point of the current Sino-US AI development competition is capital expenditure. American giants hope to continuously build and consolidate their business walls by significantly increasing AI Capex, thereby raising the entry threshold for the AI industry.

Nowadays, our leading companies are limited by the low return on investment of AI capital and do not have enough confidence to invest more capital expenditure to make up for the technological gap. However, Kuaishou Keling AI, through its product hard power, raises ROI, not only establishing a virtuous cycle for its own capital investment but also setting an example for the development of Chinese AI applications.

If AI empowerment driving business growth is the potential expectation difference for Kuaishou in the medium and short term, then the performance of Keling AI undoubtedly injects a strong boost for Kuaishou's long-term development.

In summary, through Kuaishou's interim report, we can clearly see that AI is not just a distant future vision. On a mature platform, AI technology can already deeply penetrate every link of connecting users and content, empowering merchants and supply, becoming a key engine driving core business growth.

This article is based on public information and is for information exchange only and does not constitute any investment advice. $KUAISHOU-W(01024.HK)

### 相關股票

- [Alphabet (GOOGL.US)](https://longbridge.com/zh-HK/quote/GOOGL.US.md)
- [Alphabet - C (GOOG.US)](https://longbridge.com/zh-HK/quote/GOOG.US.md)
- [Meta Platforms (META.US)](https://longbridge.com/zh-HK/quote/META.US.md)
- [KUAISHOU-WR (81024.HK)](https://longbridge.com/zh-HK/quote/81024.HK.md)
- [KUAISHOU-W (01024.HK)](https://longbridge.com/zh-HK/quote/01024.HK.md)