HappyHorse made a surprising comeback, marking the first major move in Alibaba's strategic realignment

By / Jinye

Source / Node AI

Recently, the enigmatic large video model HappyHorse has been generating significant buzz within tech circles

Yesterday, HappyHorse was officially claimed by Alibaba, originating from ATH (Alibaba Token Hub), a newly established AI core business unit

Previous analyses have suggested that 'HappyHorse' is an Alibaba product

NodeAI discovered two particularly solid pieces of evidence: First, HappyHorse’s technical approach closely resembles the path taken by Alibaba’s Tongyi Lab. Second, Zhang Di, known as the 'father of KeLing,' reportedly completed this product in just five months after returning to Alibaba at the end of 2025. However, there are also reports suggesting the product was actually developed by Zheng Bo's team at ATH rather than Zhang Di’s. Zheng Bo, Vice President of Alibaba, previously led Taobao’s search and recommendation algorithms, served as CTO of Alimama, and headed algorithm technology for Taobao and Tmall, focusing on large models, multimodality, and decision intelligence

Regardless of who specifically developed it, one conclusion is clear: the ATH division is now getting serious

Two strategic shifts, executed in quick succession

On March 16, Alibaba established the ATH division with a core mission summarized in nine words: create Token, distribute Token, apply Token

Just three weeks later, Alibaba made another move, establishing a Group Technology Committee focused on top-level design and resource coordination within its technological framework

From an outside perspective:

The first adjustment was to separate the AI business and address the question of 'who leads.'

The second adjustment broke down departmental barriers to solve the issue of 'how to cooperate.'

The group's Technology Committee is led by Wu Yongming as the head, with Zhou Jingren, Wu Zemin, and Li Feifei as members. Among them, Zhou Jingren serves as the Chief AI Architect and leads the Tongyi Large Model Division, Li Feifei is responsible for Alibaba Cloud technology and AI cloud infrastructure construction, while Wu Zemin acts as the convener of the Technology Committee and focuses on the group’s CTO responsibilities.

The roles of these three individuals are worth examining closely.

Zhou Jingren focuses on models,sending a clear signal that Alibaba is taking control of model sovereignty.

Currently, the Tongyi large model serves as the brain of Alibaba’s AI lineup. From Qwen 1.0 to 3.6 Plus, Qwen has entered the top tier globally. In terms of open-source ecosystems, Alibaba’s Tongyi has established a significant global leading position. As of the report release, Tongyi has cumulatively open-sourced over 300 models, with downloads exceeding 600 million globally, and 170,000 derivative models created, all ranking first globally. At the same time, more than one million enterprise customers have integrated Tongyi’s large model, expanding its B-end ecosystem coverage and forming scalable monetization capabilities.

Li Feifei (Feidao) manages computing power, which may signal to everyone that Alibaba intends to consolidate its cloud and AI operations.

Li Feifei’s appointment is not just a simple title addition; he now shoulders the dual role of Alibaba Cloud CTO and head of AI infrastructure. Observing industry trends, over the past two years, all major cloud players have been integrating AI capabilities into their clouds. AWS has Bedrock and Claude, Azure has the full suite of OpenAI tools, and Google Cloud offers Gemini. Li Feifei’s “dual identity” essentially means one thing: Alibaba Cloud will become an AI cloud going forward, and AI capability will be the moat of the cloud. The days of having two separate teams operating independently are over.

An industry insider told Node AI that in 2025, people were still debating whether model vendors should enter the cloud business and whether cloud vendors should develop models. By 2026, the answer was clear: the pairing of 'model vendors + cloud vendors' is becoming the industry standard.

Wu Zemin is responsible for the AI inference platform and business implementation. This business is simple to understand but difficult to execute, as it determines whether investments can truly be recouped.

He needs to embed AI capabilities into all business scenarios such as Taobao, Alipay, and Xianyu. Whether it can make money depends on his execution.

In fact, before this reorganization, Alibaba's AI division had experienced significant turmoil. In early March, the Tongyi Lab planned to spin off the Qwen team, and Lin Junye, a core member of the original Qwen team, suddenly resigned. Wu Yongming held an emergency meeting with top executives to stabilize morale.

This incident served as a wake-up call for Alibaba: the AI battle among tech giants isn't about individual business units fighting alone; it's about group-wide collaboration. Compared to the previous state where various teams worked in silos and resources were redundantly invested, this adjustment has at least significantly improved organizational efficiency.

What exactly does Alibaba want to do?

Node AI found that between these two reorganizations, the Tongyi Lab densely released three flagship models within four days: Qwen3.5-Omni multimodal interaction model, Wan2.7-Image visual generation model, and Qwen3.6-Plus large language model, covering three key capability areas: multimodal understanding, image generation, and programming agents.

So, don’t just view the viral HappyHorse as a simple update to a video-focused large model—this is Alibaba’s first “show of strength” after its organizational restructuring.

What Alibaba most wants to achieve is to accelerate the commercial implementation of AI.

Over the past few years, Alibaba's biggest problem hasn’t been a lack of funding, talent, hardware, computing power, or technical reserves, nor even a lack of application scenarios. Its real issue has been the sheer size of the group, with too many businesses and overly long decision-making chains, leading to its technical capabilities being scattered for a long time.

Each business is strong, and every team has achieved results, but when put together, it's hard to form a truly meaningful corporate-level technological campaign. Cloud is one area, e-commerce another, local life yet another, and then there's DAMO and Tongyi, often resulting in localized prosperity but overall sluggishness.

At this stage of AI, the biggest fear is sluggishness.

Because this round of competition is different from the mobile internet era. Mobile internet was about products, traffic, and channels—being a bit slower or burning more cash still left room for catch-up. In this round of large models, it's about computing power allocation, model iteration, engineering systems, data loops, and commercial implementation. A failure in any link directly drags down the entire operation. More brutally, there aren't many spectator seats this time around. Once leading companies establish technological platforms and ecosystem inertia, latecomers face extremely high costs to catch up.

In the past couple of years, the core of the large model competition was parameters and technical metrics; whoever had the stronger lab held the话语权 (note: omitted term per guidelines). But now, the competition has shifted towards the ability to deeply integrate model capabilities with business scenarios to achieve a closed loop between R&D and commercialization. Organizational synergy is the key to unlocking this closed loop.

Take a look at industry peers:

Tencent dissolved its AI Lab, merging its resources into the HunYuan large model, leveraging WeChat's 1.2 billion monthly active users to drive AI adoption in social scenarios.

ByteDance’s AI assistant 'DouBao' has surpassed 100 million monthly active users. Backed by the traffic pools of Douyin and Toutiao, DouBao quickly transformed AI from 'a geek's tool' into 'a companion for the masses.'

In reality, everyone is studying how to create a commercial closed loop within their own domains.

Not long ago, Alibaba CEO Wu Yongming set a hardcore financial target: In the next five years, cloud and AI commercial revenue will exceed $100 billion annually. However, in the fiscal year 2026, Alibaba Cloud's external commercial revenue barely surpassed 100 billion RMB. From 100 billion RMB to $100 billion, this goal means Alibaba Cloud’s revenue will need to grow sevenfold in five years, achieving over 40% annual growth. If achieved, this would be akin to creating another Alibaba.

Therefore, Alibaba must use organizational transformation to accelerate growth; otherwise, achieving this goal will be very challenging.

Alibaba's third attempt at platformization

More than two decades ago, Alibaba platformized its merchant base, selling access to transaction opportunities — the essential utilities of the e-commerce era. A decade ago, Alibaba turned computing power into a platform, with Alibaba Cloud becoming the utility provider for cloud migration. Today, as models and computing power become increasingly commoditized, all AI services will be billed based on Token consumption, just like utilities such as water, electricity, and gas. The ability to generate and invoke Tokens efficiently and at low cost has now become the new competitive focal point.

In the short term, Alibaba’s technical resources will rapidly consolidate around AI.Previously scattered R&D capabilities across various business units will now be centrally managed and coordinated. The message being sent externally is straightforward: models will be updated more quickly, cloud products will become more capable, and the pace of inference services and practical implementation will significantly accelerate.

In the medium term, internal organizational power dynamics are being reshaped.The influence of technical teams will continue to rise, while business teams that fail to capitalize on AI advancements risk declining in stature. This is harsh but also normal; once a company enters a technology-driven cycle, resources naturally flow toward departments closest to the core engine. In the past, traffic-focused departments held sway; later, it was platform-focused departments. Now, it's AI infrastructure and model platforms taking center stage.

In the long term, e-commerce remains Alibaba’s foundation, cloud computing represented its growth potential over the past decade, and AI will determine whether Alibaba can stay at the forefront for the next ten years.If the Tongyi system truly succeeds, Alibaba will evolve from a traditional internet platform into a comprehensive AI platform spanning models, cloud services, applications, and enterprise solutions. Once this position is secured, valuation logic, market narratives, and talent attraction will all be rewritten.

Of course, the biggest risk lies not in the technology itself but in organizational execution. Strategic insights are not lacking among major tech companies; what’s missing is the ability to turn corporate consensus into unified action. Currently, Alibaba’s Technical Committee has been established, and ATH’s strategic positioning has been elevated. However, if departmental silos persist, resource allocation continues to waver, and each business unit operates independently, this adjustment will amount to yet another well-intentioned but ultimately ineffective reform.

That said, Alibaba’s strategy is clear: when Tokens become a universal foundational resource, whoever achieves scale and high distribution efficiency will control pricing power. This ‘utility’ business — from commodities to computing power and now to Tokens — evolves in form but remains rooted in the fundamentals of platformization. This battle offers no retreat.

*The cover image is generated by AI.

7235 Views