Author: Chang Yuan Editor: Zhongdianjun

Over the past year, AI Coding has emerged as the most promising sector within the AI industry.

From Cursor and GitHub Copilot to code agents like Claude Code, their value proposition is clear: developers write code daily, and enterprises deliver projects on a regular basis. If AI can reduce repetitive coding tasks, minimize bugs, and save countless hours of debugging, its value becomes immediately quantifiable.

Now, another field—once considered too costly, too flashy, and too challenging to monetize—is showing signs of following a similar path. It could very well be the next sector to achieve a self-sustaining business cycle.

This field is AI video generation.

The latest milestone in this space is Kling AI. According to public reports, Kuaishou is exploring asset restructuring and external financing options for Kling AI, with a target valuation reportedly reaching as high as $20 billion. Kuaishou later clarified in a Hong Kong Stock Exchange announcement that the company is evaluating several potential transaction plans, emphasizing that discussions are still in the preliminary stages and no final agreement has been signed.

Capital markets are now reevaluating video models through the lens of an AI-native company, rather than merely as an internal tool for a short-video platform.

More critically, video models have already established a range of monetizable use cases.

Kuaishou's 2025 financial report revealed that Kling AI underwent multiple rounds of model upgrades in the fourth quarter of 2025, focusing on enhancing model capabilities, improving product experience, and advancing commercialization. Kuaishou also disclosed that AIGC marketing materials drove total client spending on online marketing services to RMB 4 billion in the fourth quarter.

As of December 2025, Kling AI had served over 60 million creators globally, generated more than 600 million videos, and forged partnerships with over 30,000 enterprise users. In December 2025 alone, its monthly operating revenue exceeded $20 million, corresponding to an annualized revenue run rate of $240 million.

Video models have evolved from the demo stage to fully billable products.

Historically, discussions about video models centered on consistency, camera movement, duration, and image quality. Now, revenue metrics must also be considered—where the money comes from and whether it can keep flowing. This is the central question this article aims to address: After AI Coding, has the video model achieved a similar self-sustaining business cycle?

Our assessment: Video models have completed a self-sustaining business cycle but have not yet achieved a closed profit loop.

  Key Players: User Willingness to Pay Trumps Data Competition and Leaderboards

If AI video generation is viewed as a poker table, the list of players is straightforward: Kling, Seedance, Veo, Runway, Vidu, Conch, Tongyi Wanxiang, Pony Dash, Hunyuan Video, Sora...

Based on the current competitive landscape, large corporations are building barriers through data and revenue growth, while startups are relentlessly iterating in the gaps.

AI video players can be broadly categorized into four types:

The first category consists of short-video platform players, represented by ByteDance's Seedance and Kuaishou's Kling.

These companies possess not just models but also vast video data, content understanding expertise, creator ecosystems, advertising systems, and monetization channels. They don't need to educate the market from scratch because users are already producing, consuming, and placing videos on their platforms.

Kling's strengths lie in its early start, rapid productization, and stronger global creator mindshare. It has evolved from an internal Kuaishou tool into an AI video infrastructure for creators, enterprises, and developers. If Kling becomes more independent, its potential will extend beyond the Kuaishou app to broader content production chains, including advertising, e-commerce, short dramas, gaming, animation, and brand marketing. However, the challenge remains significant: video inference costs are higher than those for text, images, or code. The true difficulty lies in whether revenue can cover inference costs, R&D investment, and customer acquisition costs—a challenge faced by the entire industry.

Seedance 2.0 represents a more aggressive approach. Backed by ByteDance's Douyin, Jianying, Jiying, Giant Engine, and overseas content ecosystems, it naturally integrates content production, editing, distribution, and placement. ByteDance itself understands video traffic and monetization better than anyone. In February of this year, Seedance 2.0 took the market by storm, with comic and short-drama companies flocking to it. By late March, ByteDance's comic drama daily spending exceeded RMB 70 million, surpassing live-action short dramas for the first time.

An insider at Seedance told us: Currently, user willingness to pay is extremely high, and demand for video generation using Seedance 2.0 far outstrips supply. Due to computational resource shortages, the platform has implemented queuing rules. Some short-drama clients are even willing to sign annual contracts worth tens or even hundreds of millions of yuan in guaranteed spending to secure higher queue priority.

This indicates that AI video monetization is shifting from pure consumer (ToC) subscriptions (similar to the large language model's "membership + credit consumption" mechanism) to a significant enterprise (ToB) focus. Currently, the highest demand comes from film and television companies (short dramas/comic dramas/animation), advertising and marketing firms, and internet giants, which are willing to pay for stable production capacity. Billing models have evolved directly into "package prepayments or postpayments (including API/token and credit usage)" for large enterprise clients. The explosive success of Seedance 2.0 is the result of synergies among platform capabilities, data capabilities, engineering capabilities, and commercial scenarios.

The second category consists of professional creative tool players, represented by Runway, Vidu, and Conch AI.

Runway has long positioned AI video as a tool for professional creators, emphasizing controllability, camera language, character consistency, and workflows. For professional users, a single stunning video is not enough; the real challenge is maintaining consistency for the same characters, objects, and styles across shots.

Vidu's strengths lie more in reference-based video generation and industry-specific applications. In Q2, Vidu's Reference Gen Pro emphasized that "everything can be referenced," aiming to reduce reliance on post-production fixes for characters, objects, and styles. In Q3, it focused on 16-second audio-video generation from text and images. Its enterprise demand is concentrated in comic dramas, short-drama animation, e-commerce marketing, and cultural tourism industries.

Conch AI has taken a differentiated route. Leveraging MiniMax's full-modality matrix of text, voice, music, and video, Conch AI upgraded its Media Agent, building its core competitive barrier on nuanced capture of characters' micro-expressions, natural transitions of dynamic emotions, and one-click film production with synchronized audio and visuals. For film directors and short-film creators, physical consistency is just the baseline; what truly determines film quality is characters' "acting" and authentic emotional tension. By addressing this professional need, as of the end of 2025, Conch AI's video model had not only helped global creators generate over 600 million videos but also seen its open platform and service revenue for enterprise clients surge by 197.8% year-over-year, proving its strength in content production.

Large platforms excel at turning video generation into foundational capabilities and traffic entry points, but professional creative tools can offer finer-grained solutions in controllability, team collaboration, asset management, copyright marking, API access, and industry templates. They may not have the most data or traffic, but by becoming the production system for specific client segments, they can thrive.

The third category consists of cloud computing and ecosystem players, represented by Google Veo, Alibaba's Tongyi Wanxiang, HappyHorse, and Tencent's Hunyuan Video.

These companies may not develop AI video as a standalone app but embed it into cloud services, advertising systems, office tools, and developer APIs. For example, Google's Veo 4 emphasizes native audio generation and physical realism, connecting with Gemini, YouTube, advertising, and cloud APIs. These players have abundant resources and broad ecosystems, but video models are just one piece of their multimodal capability puzzle, not necessarily the sole focus.

The fourth category consists of short-lived players, with Sora as the prime example.

Sora once dominated headlines and was hailed as the ChatGPT of video, but OpenAI officially discontinued Sora's web and app experiences on April 26 this year, with API access ending on September 24.

The retreat of the most representative AI company underscores a harsh reality: Video generation monetization cannot rely solely on technical spectacle and social buzz. Customer willingness to pay matters more than model parameter competition and leaderboard rankings.

According to industry information, among domestic players, Seedance now dominates the market with over 80% of daily spending, followed by Kling at around 14%, Wanxiang 2.7 at about 4%, and HappyHorse at less than 1%. Seedance leads by a wide margin, Kling holds a solid second-tier position, and Alibaba's dual models combined account for about 5%, ranking third.

  Seedance and Kling Lead: Is the Video Model Landscape Repeating the Short Video Pattern?

For a long time, Kling has been synonymous with AI video in China. It launched early, generated significant buzz, and achieved commercial revenue sooner. When Seedance 2.0 suddenly gained intense attention, a natural question arose:

Was this a coincidental overtaking, or an inevitable outcome?

In any single round of model performance, who leads by a few months or performs better in a specific scenario may involve phased coincidences. But in the long run, what matters is the comprehensive capability to convert data, models, engineering, and scenarios into commercial output.

This is where Seedance's pursuit of Kling becomes most noteworthy.

Video models differ from text models. While high-quality text corpora are important, video models inherently rely more on multimodal data: visuals, camera movement, actions, pacing, audio, subtitles, user feedback, watch completion rates, engagement rates, and advertising effectiveness.

Short-video platforms happen to control all these elements.

ByteDance and Kuaishou's advantages extend beyond merely owning vast video libraries. More importantly, they understand which videos get watched to the end, which openings retain viewers, which pacing suits information feeds, which materials drive advertising spending, and which content converts into transactions.

These insights may not all become training data directly, but they inform product design, data curation, model evaluation, and monetization paths.

This explains why video generation competition may resemble short-video platform rivalries more than general-purpose large model contests focused solely on parameters, leaderboards, and open-source ecosystems. More data leads to faster feedback; faster feedback enables more precise model and product iterations; more precise products increase customer willingness to pay; more customers, in turn, generate more real-world usage data.

This is also where video models most resemble AI Coding.

AI Coding achieved a self-sustaining business cycle because it integrated into developers' daily workflows. It was continuously invoked, its effects continuously verified, and its value continuously billed.

For AI video to achieve a self-sustaining business cycle, it must similarly become an indispensable production tool for content creators, advertising placers, e-commerce operators, and short-drama teams—not just a toy for occasionally creating viral videos.

Thus, the industry has entered a new phase where the focus is on who can provide stable output capacity. Next, the competition will shift to unit costs, delivery stability, industry templates, copyright compliance, team workflows, and key account lock-in capabilities.

The competitive advantages for video models include at least the following:

Data. Especially video content, user feedback, and commercial placement data.

Model capabilities. Including image quality, motion, physics, camera movement, multi-shot coordination, audio-video synchronization, and prompt understanding.

Engineering efficiency. Lower costs, faster generation, higher concurrency, and more stable delivery.

Product workflows. Including first/last frames, reference images, character consistency, partial redrawing, camera control, team collaboration, asset management, and APIs.

Client budgets. Only those who can tap into advertising, e-commerce, short dramas, gaming, and brand marketing—industries with sustained spending—can establish a harder commercial loop.

Leading in a single model dimension is important but insufficient. Especially in early commercialization stages, a 90-point solution with ten times higher costs may not outperform a 75-point solution that is stable, affordable, and scalable.

An industry-wide concern is whether, with ByteDance and Kuaishou already the top two, the video generation landscape is already set.

This judgment holds some merit because video generation and short-video platforms are naturally coupled:

Data coupling: Platforms own vast videos and feedback.

Scenario coupling: E-commerce, livestreaming, and advertising are the easiest monetization grounds.

Commercial coupling: Advertisers and merchants already spend on these platforms; efficiency gains can directly cut into their budgets.

Distribution coupling: Whether generated content gets seen and converts determines client renewal willingness.

From this perspective, ByteDance and Kuaishou indeed seem like natural winners. But this doesn't mean other players have no opportunities.

Because the AI video market is vast enough for diverse players to find their niches:

If understood as who generates more in-feed ads, platform players have a massive advantage. But if seen as the infrastructure for all future video production workflows, professional tools, cloud vendors, vertical models, and industry solutions still have room.

Other players can leverage their unique strengths to carve out distinct niches:

Professional controllability. The film, gaming, and brand advertising industries demand utmost precision and consistency. Characters must maintain visual coherence, and products must be rendered without distortion. Those who can stabilize professional workflows stand to capture lucrative markets.

Cost-efficiency and APIs. SaaS companies and design platforms may not require the highest image quality but urgently need affordable, stable, and high-concurrency APIs to support their operations.

Global market expansion. Chinese video models are now on par with global leaders. For both established corporations and startups, "going global" is a top priority, with Japan and South Korea emerging as key markets for overseas revenue generation.

The AI video industry is likely to evolve into a two-tiered ecosystem: Upper-tier platform giants will dominate with massive data volumes and advertising budgets, serving as foundational productivity tools. Meanwhile, lower-tier tool and industry-specific players will specialize in professional content creation, overseas market nodes, and enterprise workflows.

This mirrors the trajectory of AI coding, where general-purpose large models provide the foundation, Cursor emerges as the workflow tool, and enterprises integrate solutions based on their security needs. The true value lies in the synergy of "model + tool + workflow + enterprise scenarios."

AI video is poised to follow a similar path. While foundational models grow more powerful, commercial applications will materialize closer to end clients: advertising placement systems, short-drama production pipelines, e-commerce content factories, game asset generation, brand content management, and creator workstations.

Three Key Hurdles to Closed-Loop Profitability for Video Models

A common misconception about AI video monetization is equating it solely with consumer-side subscription models, leading to an overemphasis on metrics like active user counts.

While creators' willingness to pay for watermark removal and higher usage quotas is important, it barely scratches the surface. The video model's business model is more akin to a hybrid of ad technology, content supply chains, and production tools.

Consider an e-commerce merchant who previously invested significant time and money in short-video production and editing. If AI can generate dozens of content variants from product images and selling points, then rapidly filter them based on conversion rates, it effectively becomes part of the advertising machine. The same logic applies to short dramas and casual games. AI video may not immediately replace entire short dramas but can first replace trailers, teaser clips, and plot test materials.

Thus, the first area where AI videos successfully achieve a closed business loop is likely not the film industry but rather the production of repetitive, mass-produced, and performance-evaluated commercial video content. This may lack the allure of Hollywood glamour, but it is grounded in reality. The film industry pursues artistic excellence, while advertising focuses on ROI. The former fuels imagination, while the latter drives cash flow.

For C-end users and creators, subscriptions offer access to usage quotas, processing speed, clarity, and advanced features. For enterprise SaaS, services cater to clients in advertising, e-commerce, branding, MCNs, short dramas, gaming, and more. APIs and MaaS enable seamless integration with third-party applications, marketing platforms, and design tools. Internal platform revenue growth involves incorporating AIGC content into advertising, e-commerce operations, and live-stream interactions, ultimately boosting advertiser spending and merchant operational efficiency.

Among these monetization paths, standalone C-end subscriptions are often overestimated, while internal platform revenue growth warrants the most attention. The needs of ordinary consumers are intermittent, and the true heavy users are content creators, advertising professionals, and operations teams at major internet companies. Those requiring mass distribution, stable output, and willing to pay for priority computing access are the more reliable revenue sources.

The willingness of short-drama clients to sign multi-million-yuan guaranteed consumption contracts with Seedance underscores that video models have become the gateway to content industry production capacity.

Of course, for video models to achieve true profitability, several challenges must be overcome:

Controllability. Commercial delivery cannot rely on chance; advertising clients and film/TV teams demand predictable results.

Cost management. Is AI more cost-effective than outsourcing? For startups, balancing the high computational costs of generation with sustainable, healthy revenue is a survival imperative.

Compliance and risk mitigation. Videos are more sensitive than text, involving issues like portrait rights and deepfake risks. Currently, the domestic industry pays relatively little attention to data security and copyright concerns in model training. However, as commercialization deepens, this gray area urgently requires regulation and strict oversight. The closer to core businesses and major IPs, the higher the compliance costs will be.

Conclusion

Returning to the original question: After AI coding, are video models successfully achieving a closed business loop?

The answer is yes—but only in the narrow yet high-potential field of high-frequency commercial video production. Its first major revenue stream comes from scenarios like short dramas, advertising, marketing, and e-commerce, which have sustained content demand and are willing to pay for efficiency.

The next challenge is ensuring clients engage with the technology daily, using it consistently, and paying for its value. This is a path validated by AI coding and remains the key to whether video models can achieve sustainable profitability.