Produced by | He Xi

Layout by | Ye Yuan

This spring, the global AI community has experienced an unprecedented upheaval.

On March 24, OpenAI shut down Sora. The 'multimodal marvel' that stunned the world two years ago and surpassed 1 million downloads in just five days came to an abrupt end. Meanwhile, ByteDance's Seedance defied the industry's price war by raising prices three times in a month, adjusting its membership system to charge approximately 15 yuan for a 15-second video—equivalent to 'one yuan per second.' As users complained about the price hike, a dark horse named HappyHorse suddenly topped authoritative evaluation rankings, surpassing Seedance 2.0 and Kling 3.0, before being officially 'claimed' by Alibaba.

These three events, seemingly independent, point to the same signal: the 'technological fireworks' of AI video are fading, and the 'value competition' has begun.

Below, I discuss these three anomalies—Sora's shutdown, Seedance's price hike, and HappyHorse's rise—and share my personal observations on the future of AI video.

01

Three Events, One Inflection Point

Let's start with the internal logic of these three events.

First, Sora's shutdown. This product, which made a stunning debut two years ago and surpassed 1 million downloads in five days, incurred daily inference costs of $15 million just for regular consumer operations, according to SemiAnalysis. Annual operating costs exceeded $5 billion, while its cumulative revenue over its entire lifecycle was only about $2.1 million. With such a dismal return on investment, shutdown was almost inevitable.

Sora's exit leaves a profound lesson: Technical brilliance alone cannot sustain a business. Only those who can transform AI video from a 'money-burning toy' into a 'cost-effective tool' will survive.

Next, Seedance's price hike. In early 2026, major AI players were embroiled in a price war, desperate to acquire users even at a loss. However, ByteDance's Seedance 2.0 went against the tide, raising prices three times in a month. According to Orient Securities, the price increases for Seedance 1.5 Pro and 2.0 standard versions were 49% and 99%, respectively. Generating a 15-second video now requires 210 credits, roughly 15 yuan—'one yuan per second.'

Seedance's confidence in raising prices stems from its technical accumulate (accumulation). It supports four modalities—text, image, audio, and video—and can generate high-quality, multi-angle videos up to 15 seconds long, demonstrating stability in complex interactive logic and high-dynamic motion scenes, with preliminary industrial production capabilities. However, the price hike sparked widespread dissatisfaction among creators, with some showing over 300,000 people still queuing at 3 a.m., expecting waits exceeding two hours.

Finally, HappyHorse's rise. Amid user discontent over Seedance's price hike, an anonymous model quietly topped Artificial Analysis's Video Arena rankings, surpassing Seedance 2.0 and Kling 3.0 in both text-to-video and image-to-video (audio-free) categories.

On April 10, Alibaba officially 'claimed' HappyHorse. However, contrary to early rumors of 'fully open-source and commercially usable,' Alibaba later clarified that HappyHorse would not be open-sourced and would instead offer commercial services via API through the BaiLian platform.

Viewed together, these three events reveal a clear trend: The competitive logic of AI video is shifting from 'who has higher technical parameters' to 'who has a more sustainable business model.' Sora's exit, Seedance's price hike, and HappyHorse's comeback collectively proclamation (declare) the end of an old era and the beginning of a new one.

02

The Clash of Titans: Alibaba vs. ByteDance's Strategic Divide

After discussing the inflection point, let's examine the core contradiction in the AI video race.

If the AI video track (race) is a chessboard, the main competition now lies between Alibaba and ByteDance. This is not merely a technical duel but a clash of two business philosophies: 'open-source ecosystem' versus 'closed-source productization.'

Technical Architectural Divide

HappyHorse-1.0 is a 15-billion-parameter video large model adopting a 40-layer unified self-attention Transformer architecture. It is the world's first open-source video model natively supporting joint audio-video generation (model weights open-sourced, but commercial use requires API access). 'Natively joint audio-video generation' means it can produce videos with synchronized audio in one go, rather than a 'patchwork' approach of generating visuals first and adding audio later, offering a clear advantage in user perception.

Seedance 2.0, on the other hand, employs a dual-branch Diffusion Transformer architecture, validated by large-scale commercial deployment, offering greater maturity in stability and controllability. Its biggest strength lies in engineering capabilities—stable performance in complex interactive logic and high-dynamic motion scenes, with preliminary industrial production capabilities.

Each architecture has its pros and cons. The single-stream Transformer is more streamlined but less mature, while the dual-branch Diffusion is more engineering-proven but may lack flexibility in certain scenarios.

Open-Source vs. Closed-Source Business Logic

This is the most striking difference between the two companies' strategies.

Alibaba's approach with HappyHorse is clear: open-source model weights, monetized via API through the BaiLian platform. Strategically, Alibaba aims to be the 'Android' of AI video—building an ecosystem foundation for developers to build applications. With API access opened on April 30, Alibaba is prepared for 'open-source for ecosystem, closed-source for commerce.'

ByteDance, meanwhile, pursues a closed-source productization route with a full-stack layout: Jimeng App (consumer end), Jianying (tool end), AI MediaKit (enterprise end), and Volcano Engine API. This forms a complete commercialization closed loop (loop)—users engage, use, and pay all within ByteDance's ecosystem. Strategically, ByteDance aims to be the 'Apple' of AI video—software-hardware integration with a closed experience loop.

User Sentiment and Market Window

Seedance's consecutive price hikes created a rare 'window of expectation' for HappyHorse. Some creators argue that with intense competition among domestic large models, they are not worried about lacking alternatives to Seedance. Seedance's decision to accelerate monetization before the end of its technological dividend period suggests a lack of absolute confidence in sustaining its lead.

This judgment is not unfounded. With HappyHorse topping evaluation rankings as an open-source model, the cost of user migration is nearly zero. Industry analysis points out that users will only pay for better and cheaper AI models—when Seedance 2.0 was superior, creators flocked to ByteDance; now that HappyHorse excels, customers will naturally divert or switch camps. Indeed, according to Jiemian News, Alibaba Cloud's frontline sales have begun actively poaching AI video clients from Volcano Engine.

Ecosystem Synergy: Breadth vs. Depth

This is the fundamental difference between the two companies.

ByteDance's strength lies in 'full-stack distribution.' Douyin, Jimeng, and Jianying form a complete closed loop (loop) from content production to distribution. Users generating videos with Seedance can publish them to Douyin with one click, creating a 'production-distribution-monetization' loop that is hard for others to replicate.

Alibaba's strength lies in 'deep integration with e-commerce scenarios.' Scenarios like Taobao and Tmall product videos and advertising materials have substantial demand for AI video. If HappyHorse can succeed in these scenarios, it will secure a stable revenue stream. Meanwhile, its open-source strategy helps attract long-tail developers, accelerating ecosystem growth.

Both routes have their merits, but victory hinges not on technological flair but on who can faster achieve a commercial closed loop (loop).

03

Kuaishou and Tencent: Differentiated Approaches

After discussing Alibaba and ByteDance's duel, let's examine the other two major players in AI video—Kuaishou and Tencent—and their differentiated strategies.

Beyond the 'duel' between Alibaba and ByteDance, Kuaishou and Tencent are also significant players in the AI video race. While their technical evaluation rankings currently lag behind the top two, their differentiated strategies have allowed them to establish irreplaceable advantages in niche areas.

Kuaishou Kling: Pioneering Commercialization

Kling boasts the most robust monetization among the 'Big Four.' According to exclusive research by LatePost, its 2025 annual revenue is projected to reach $140 million, far exceeding the initial target of $60 million. Monthly revenue surpassed $20 million in December 2025, with daily average revenue rising another 30% in January. Nearly 70% of its revenue comes from professional subscription members, and it topped download charts in 40 countries, including Brazil and Russia. As of January 2026, Kling AI's monthly active users exceeded 12 million, with paid users on the App end surging 350% month-over-month.

Kling's strength lies in its 'creator ecosystem.' It better understands content creators' needs—strong cinematic narrative sense and film-grade quality, suitable for producing short films. The viral short film 'Paper Phone' during this year's Qingming Festival was created using Kling. This reputation among professional creators is Kling's most valuable asset.

Tencent Hunyuan: The Lightweight, Open-Source Approach

Tencent has taken a completely different path. In November 2025, it released HunyuanVideo 1.5, a lightweight open-source video generation model with only 830 million parameters, capable of generating high-quality videos on consumer-grade graphics cards (e.g., RTX 4090).

The brilliance of this strategy lies in 'lowering barriers.' While other players compete on parameter scale and generation quality, Tencent chooses to 'enable more people to use it.' Critically, Tencent owns a vast library of online literature IP through China Literature and content platforms like 'Huolong Manju,' forming a 'IP→content→AI generation' content closed loop (loop).

Tencent Hunyuan aims to lower video creation barriers and attract long-tail users. While this may not yield significant commercial revenue shortly, it is a crucial step in laying the groundwork for the 'era of universal creation.'

Epilogue

After discussing the competitive landscape, let me share my insights on the future of AI video—or rather, the three hurdles AI video must overcome.

The first is technological scalability. Inference costs must continue to decline. Seedance has reduced video generation costs to 'one yuan per second,' but this is still far from large-scale commercialization. HappyHorse's '8-step denoising' technology significantly lowers generation costs, but whether it can undercut Seedance remains to be seen.

The second is commercial sustainability. Whether through subscriptions, API calls, or open-source value-added models, gross margins must turn positive. Sora's fate already shows that unsustainable business models cannot last.

The third is regulatory compliance. Risks like AI face-swapping and voice cloning are becoming 'gray rhinos.' Volcano Engine has established a full-chain portrait rights and copyright protection mechanism for Seedance, covering pre-creation authorization management, in-creation content generation, and post-creation output review. The platform offers over 10,000 high-quality virtual human images spanning different age groups, professional backgrounds, and styles. This is a common challenge all AI video players must address.

Sora's shutdown, Seedance's price hike, and HappyHorse's rise collectively proclamation (declare) the end of one era and the beginning of another.

The competition in AI video is no longer about 'who can generate cooler videos' but 'who can make AI video a productive tool at lower cost, higher efficiency, and greater compliance.'

In this paradigm shift from 'technological sprint' to 'commercial closed loop (loop),' the true winner will be the enterprise that first establishes a virtuous cycle among 'technology-scenario-commerce-compliance.'

Alibaba has e-commerce scenarios and an ecosystem foundation; ByteDance has traffic distribution and productization capabilities; Kuaishou has a creator base and monetization experience; Tencent has IP content and lightweight open-source strategies. Each has strengths and weaknesses.

As technological gaps narrow, the decisive factor will shift from 'speed' to 'endurance.' Whoever finds a sustainable balance among computing costs, commercialization, and regulatory compliance will dominate the next decade of AI video.