Introduction: Five Years for Kling, Five Months for HappyHorse—The Architect Behind Kuaishou's AI Becomes Its Biggest Competitor

Alibaba has truly embraced Lei Jun's marketing approach.

Last month, a mysterious model, codenamed 'Hunter Alpha,' quietly entered the large model market, showcasing remarkable performance. It was only after its results were verified and discussions intensified that Lei Jun casually announced, "This is ours."

Now, Alibaba has adopted the same tactic.

Recently, an anonymous model named HappyHorse-1.0 silently debuted at the top of the Artificial Analysis Video Arena rankings, surpassing ByteDance's Seedance 2.0 in both text-to-video and image-to-video categories.

Subsequently, Alibaba claimed ownership of this video large model, immediately sparking widespread attention.

In the past, companies would first create hype and showcase PPTs before product launches to drive up stock prices. Now, major players have become more astute: they allow their products to anonymously prove themselves in the market before stepping forward to claim credit.

However, success ultimately depends on technical strength.

This tactic of entering the market anonymously and then revealing the identity—known as "hunger marketing" five years ago—is now a bold display of confidence. Without solid performance, anonymity risks humiliation and potential abandonment. With a decisive lead in scores, anonymity becomes a tantalizing teaser. Alibaba's audacity stems from the model's robust capabilities and the mastermind behind it—Zhang Di.

I. The Talent War for Large Models

Zhang Di, a computer science graduate with a master's degree from Shanghai Jiao Tong University, began his career at Alibaba in 2010, where he oversaw engineering architecture for Alimama's big data and machine learning for an extended period.

He later joined Kuaishou as VP of Technology, leading the foundational architecture of the Kling large model.

Under Zhang's leadership, Kling emerged as a flagship in China's AI video generation landscape. Its rise established Kuaishou's first meaningful competitive moat in AI video and laid the groundwork for its next-stage core strategy.

For reasons unknown, Zhang left Kuaishou last year. After a brief stint at Bilibili, he swiftly returned to Alibaba as head of Taotian Group's Future Life Lab. Within five months, his team developed HappyHorse-1.0.

The former "Father of Kling" now stands as one of its most formidable rivals.

The emergence of HappyHorse-1.0 underscores that the large model competition has reached a stage where "talent" determines outcomes.

Behind every recent successful large model lies a strategic "talent."

In February 2025, former Google DeepMind VP of Research Wu Yonghui joined ByteDance to lead model R&D. A year later, Seedance 2.0 launched, dominating both domestic and international markets.

By late 2025, Xiaomi's Lei Jun personally recruited Luo Fuli with a multimillion-dollar salary. Within three months, Xiaomi unveiled its large model.

Now, with Zhang Di in the mix, tech giants do not lack resources—they lack the strategists to orchestrate them.

This is no coincidence.

In traditional manufacturing, a top engineer might boost yield rates by 5%. In chipmaking, an architect can decide a product's fate. In the large model arena, the impact of elite talent is already evident.

Didn't Kling, with Zhang Di, once lead the video AI large model race? Yet two months after his departure, Seedance 2.0 stole the spotlight.

Zhang's exit is not an isolated case. In December 2025, Kuaishou VP and head of foundational and recommendation models Zhou Guorui was reported to be leaving.

Since Cheng Yixiao became chairman in October 2023, over ten Kuaishou VPs have resigned or stepped down.

Tianyancha app data reveals Kuaishou's 2025 financials were strong: RMB 142.8 billion in revenue (+12.5% YoY), RMB 20.6 billion in adjusted net profit (+16.5% YoY). Kling demonstrated robust growth, generating RMB 340 million in Q4 2025 revenue and over RMB 1 billion annually.

Kuaishou prioritizes Kling internally. At earnings calls, Cheng Yixiao stated the company would increase investment in large model R&D and applications for content and commercial ecosystems, even restoring R&D spending to ~RMB 14.5 billion after two years of criticism.

Yet compared to rivals' talent retention efforts, Kuaishou falls short amid frequent AI R&D departures.

As video large models iterate faster, elite talent's influence will grow. Algorithms can be open-sourced, computing power procured—but a mind's intuition cannot be bought or replicated.

Thus, the last laugh will belong to companies that provide the best stage for top talent, whether Kuaishou, Alibaba, or ByteDance.

In this era where "talent defines the game," Kuaishou may need to rethink its talent strategy.

Kling's fall from leader to laggard took just a year. The video generation AI race operates on monthly cycles, not yearly ones. Whether Kuaishou can keep pace depends on retaining its next Zhang Di.

II. Will HappyHorse Disrupt More Than Rankings?

If strategists represent the first layer of competition, rewriting the rules forms the second.

This year, OpenAI pivoted by shutting down Sora's standalone app. The video generation AI landscape stands at a critical juncture, with the game now hinging on "strategic depth, precision, and efficiency" rather than "first-mover advantage."

Unlike Kling or Seedance 2.0, HappyHorse-1.0 is open-source. This bold move reflects technical confidence while rewriting the foundational rules of closed-source video model ecosystems.

Seedance 2.0 and Kling-series models primarily monetize through API sales and memberships.

Seedance 2.0, for instance, raised prices three times in a month: hiking point consumption in March, eliminating old-user discounts, and launching a VIP membership tier. Users now joke that "Ji Meng" (Instant Dream) has become "Ji Gui" (Instant Expensive).

HappyHorse-1.0 chose open-source. When a decisively superior model becomes freely available, it reshapes pricing power across closed-source models.

HappyHorse-1.0's arrival signals not just a new contender but the dawn of comprehensive commercial value comparisons in video generation AI.

However, this doesn't mean HappyHorse-1.0 has checkmated its rivals.

The video generation AI arena reshuffles monthly. From Sora to Kling, then to Seedance 2.0 and HappyHorse-1.0, users have never pledged loyalty. Creators are nomadic, migrating to platforms with richer resources. When Seedance 2.0 excelled early this year, creators flocked to ByteDance's pastures. Now that HappyHorse-1.0 tops the charts, Kling and Ji Meng's flocks will inevitably split.

These AI video creators relentlessly pursue cost-efficiency and are price-sensitive. Few will pay for "second-best."

Yet HappyHorse-1.0 hasn't officially launched, and new players like Tencent's rumored "DreamNow" AIGC platform may emerge next month.

Moreover, ByteDance and Tencent boast richer application ecosystems than Alibaba.

The AI short drama market serves as a litmus test.

DataEye Research projects the 2026 web drama market to exceed RMB 24.36 billion. In March 2026 alone, AI short drama investments reached USD 1.46 billion.

On March 24, ByteDance's web drama sector saw daily ad spend surpass RMB 70 million, overtaking live-action short dramas for the first time. AI content's efficiency and cost advantages are translating into commercial dominance.

ByteDance has seized early momentum in AI short dramas with Seedance 2.0.

What about Alibaba?

Alibaba owns Youku, but it has long struggled against iQiyi and Tencent Video in long-form video and lags in short-form content. However, Alibaba's e-commerce ecosystem offers fertile ground for video AI experiments, including product showcase videos, live-streaming clips, and ad materials.

Furthermore, Alibaba's AI vision transcends short-term gains. From Alipay to QianWen to HappyHorse-1.0, it may be orchestrating a grander strategy.

Just as "Token" was recently rebranded as "token" (word elements), ordinary users increasingly recognize the value of AI's foundational resources.

Alibaba may aim to dominate the entire AI ecosystem infrastructure, akin to Alipay. As developers and enterprises integrate with HappyHorse-1.0, QianWen, and Alipay, they could ultimately flow into Alibaba Cloud's ecosystem.

Alibaba's strategy may quietly shift from "selling model APIs" to "selling foundational computing power." After all, Alibaba ranks among China's top cloud computing providers.

Like NVIDIA's real rival being the end of Moore's Law rather than AMD, or SpaceX's true challenge being gravity rather than Blue Origin, Alibaba's opponent may not be ByteDance but the question of "what defines a superior digital life."

From Alipay to QianWen to HappyHorse-1.0, Alibaba's AI playbook follows a coherent strategy. It's not scattered moves but a dual-wheel deployment around "cloud computing infrastructure + open-source model ecosystem."

Thus, the future of video generation AI promises riveting competition, with no one able to predict the final board state.

One certainty remains: internal battles define river boundaries, while external expansion opens oceans. The tech realm is vast enough for Alibaba, ByteDance, and new players alike.

Yet borders must be refined through clashes, and bottom lines tested through competition.

The next move in video generation AI has just begun.

Disclaimer: This article is based on legally disclosed corporate information and publicly available data. The author does not guarantee completeness or timeliness.

Additional Note: Stock markets involve risk. Enter cautiously. This article does not constitute investment advice; investors must conduct their own due diligence.