Summary: Following organizational restructuring, Alibaba is frequently showcasing its prowess. Can it emerge victorious in the Token war of the AI era?

Recently, Alibaba AI has been frequently demonstrating its capabilities.

At the model level, the Qwen3.5-Omni full-modal interactive model, Wan2.7-Image visual generation model, and Qwen3.6-Plus large language model were all intensively released from late March to early April.

At the application level, the natural language development application platform Meoo (Instant Comprehension), launched on April 15, can generate websites and H5 pages within a minute and deploy them with a single click.

Notably, a few days ago, Alibaba's video generation model HappyHorse-1.0 (Happy Horse) made an anonymous debut on the Video Arena list of the authoritative evaluation platform Artificial Analysis, outperforming ByteDance's Seedance2.0 and Kuaishou's Kling3.0 in both text-to-video and image-to-video categories.

Lin Junyang, the former technical lead of the Qwen large model who has since left Alibaba, openly reposted the news, stating, "Happy Horse is insanely happy."

Why has Alibaba been frequently showcasing its prowess recently? This requires a discussion of the recent turbulence Alibaba AI has experienced.

In early March, Lin Junyang, a key figure in the Qwen team, announced his departure. On the same day, Yu Bowen, the post-training lead of Qwen, also left.

Although Wu Yongming urgently convened a Qwen team meeting and issued a letter to stabilize morale, the widespread sentiment that "Qwen is nothing without its people" reflected both external and internal concerns over talent loss and R&D stability.

Faced with this situation, Alibaba initiated what can be described as the "most intensive" AI organizational transformation in its history.

On March 16, Alibaba established the Alibaba Token Hub (ATH) Business Group. On April 8, Wu Yongming issued another internal letter announcing the establishment of the Group Technical Committee and upgrading the Tongyi Large Model Business Unit.

In just three weeks, Wu Yongming completed a comprehensive centralization of Alibaba AI through these two restructurings.

Changes in a company's internal and external environment often prompt it to proactively undergo organizational transformation. At this industry turning point, with the Token economy booming, can Alibaba AI's comprehensive centralization win the Token war in the AI era?

I. Alibaba AI's 'Fiefdoms' and Wu Yongming's Intervention

After Qin unified the six states, Emperor Qin Shi Huang abolished feudalism, implemented prefectures and counties, standardized writing, and unified track gauges, ending centuries of chaos caused by feudal separatism with a centralized system.

Two thousand years later, when 'Seven Warring States' emerged within Alibaba during the AI era, Wu Yongming, wielding the CEO's authority as the 'Sword of Heaven' and organizational adjustments as a 'scalpel,' sought to unify not writing and track gauges but the production, distribution, and consumption of Tokens.

Lin Junyang's departure was merely the trigger. Why must Alibaba undergo organizational transformation?

On one hand, Alibaba AI has long been in a fragmented state.

Not to mention Lingguang and Afu under Ant Group, before the emergence of the Qianwen APP, C-end users' perception of Alibaba AI was primarily focused on Quark AI.

If this is the case for Alibaba AI's brand recognition among C-end users, the situation within Alibaba's organization is even more fragmented.

Business lines such as Quark, Taotian, and DingTalk each have their own AI research directions. The Tongyi Lab is responsible for foundational research, Alibaba Cloud promotes AI service platforms, and model R&D, platform capabilities, and business scenarios often span multiple departments.

Cross-departmental collaboration is often accompanied by high communication costs, with teams separated by 'thick barriers.'

Take the Qwen model team as an example. According to an article in LatePost, the Qwen model team formed its own Infra team last year, a task originally handled by Alibaba Cloud's AI platform PAI, which also supported Infra needs from different teams within the Tongyi Lab.

Based on his technical trend judgment, Lin Junyang believed that the pre-training, post-training, and Infra teams should be more closely integrated and communicate more effectively to resolve issues encountered in model R&D more quickly.

Alibaba AI's 'fiefdoms' stem from the company's internal 'horse racing' mechanism and the '1+6+N' organizational transformation in 2023.

At that time, Alibaba split itself into six business groups and N independent companies, attempting to activate the entrepreneurial spirit of each business unit through 'separate kitchens' and facilitate independent financing and listings.

However, this caused internal collaboration to fracture. Taotian's launch of 'Hourly Delivery' required renegotiating interface agreements with Ele.me; the price of Alibaba Cloud services purchased by the Great Entertainment division was even higher than that for external clients.

Since Wu Yongming took over as CEO, Alibaba has gradually adjusted back from the '1+6+N' model, but the 'fiefdom' situation persists.

The recently popular HappyHorse-1.0 (Happy Horse) originated from the Future Life Lab of the Taotian Group, while the Qwen team also has a text-to-image model, Qwen-image, and the Tongyi Lab has Tongyi Wanxiang (primarily for multimodal generation), indicating significant business line overlap.

In economics, there is a concept called friction cost, which refers to the additional costs incurred due to inefficient resource allocation caused by market imperfections, information asymmetry, transaction costs, and other factors in economic activities.

In a corporate context, this often manifests as endless meetings for employee communication to reduce friction costs arising from information asymmetry. Prolonged 'fiefdoms' have led to high communication costs and resource internal consumption in cross-departmental collaboration at Alibaba AI, even resulting in internal conflicts.

This is also evident from Lin Junyang's departure. According to a LatePost article, an Alibaba insider stated, "If there are dissatisfactions, they can be communicated," and one should not speak out on social media.

On the eve of Lin Junyang's departure, he only then learned that the post-training part of his team would be managed by Zhou Hao, a researcher from DeepMind. When asked by the core management, the Qwen team was told, "Zhou Hao's joining is not to replace anyone," and "therefore, there has been no communication for now."

On the other hand, Alibaba AI faces various external pressures.

At the C-end, Qianwen is under immense pressure to catch up.

According to QuestMobile data, during the Spring Festival, Alibaba spent billions of yuan in red envelope promotions for the Qianwen App, with daily active users (DAUs) soaring from around 17 million before the campaign to a peak of 73.5 million. However, after the subsidies ended, DAUs plummeted, with user retention and brand recognition falling short of expectations, and the gap with ByteDance's Doubao did not substantially narrow.

Meanwhile, the comprehensive war in the Token economy has begun, with all major players joining the fray.

ByteDance's Doubao, Tencent's Hunyuan, and Baidu's Wenxin have all transitioned from technical reserves to full-scale commercialization. While Alibaba was preoccupied with internal friction, its competitors had already completed their deployments from model R&D to commercial implementation.

ByteDance had already integrated AI Lab into Seed as a whole last year, led by Wu Yonghui; Tencent also recently dissolved AI Lab, with some personnel integrated into the Large Language Model Department, overseen by Yao Shunyu; Baidu, meanwhile, raised its AI-related revenue growth target for 2026 by 200%.

Faced with internal 'fiefdoms' and external competitive pressures, Wu Yongming initiated a 'scalpel-like' organizational transformation.

The first cut: Using Tokens as an anchor to promote Alibaba AI's unification and establish the ATH Business Group.

Wu Yongming personally led the integration of the Tongyi Lab, MaaS business line, Qianwen Business Unit, Wukong Business Unit, and AI Innovation Business Unit, achieving a comprehensive layout from foundational model R&D and model service platforms to AI applications for individuals and enterprises. The core goal is to create, distribute, and apply Tokens.

The second cut: Establishing a 'separation of powers' to break down departmental barriers and form a Technical Committee.

Wu Yongming personally served as the group leader, with Zhou Jingren stepping down as Alibaba Cloud CTO to fully oversee the Tongyi Large Model Business Unit, Li Feifei taking over as Alibaba Cloud CTO to handle AI cloud infrastructure, and Wu Zeming serving as Group CTO to oversee the technology platform and AI platform construction.

This structure consolidated Alibaba AI's previously dispersed routes of 'cloud, large models, and business technology' into a unified 'wartime tackle key problems (assault) mode,' breaking down the internal friction where 'model teams complained about insufficient computing power, cloud teams complained about overly heavy models, and platform teams believed adaptations were subpar.'

With these two cuts, Wu Yongming completed a comprehensive centralization of Alibaba AI. With the Group CEO directly leading the team, Alibaba AI eliminated individual heroism and made the organization the core competitive force in the AI era.

II. Can Alibaba AI's Full-Stack Path Succeed in the Token Economy After Centralization?

The organizational structure is now in place, but can Alibaba's 'centralized' organization win the comprehensive war among major players in the Token economy era?

According to data from the National Data Bureau, in March 2026, China's daily average Token usage surpassed 140 trillion, a more than 1,000-fold increase from 100 billion in early 2024. The industrial turning point of the Token economy has arrived.

At the GTC conference in March this year, Jensen Huang pointed out that the inflection point for AI inference has arrived, and Tokens are the new generation of commodities. Large model capabilities are transforming from 'technical parameters' into 'measurable productivity.'

The turning point for 'Tokens' as a new resource has arrived.

At Alibaba's Q3 2026 earnings call on March 19, Wu Yongming announced the financial goals of Alibaba Group's AI strategy: Within five years, cloud and AI commercialization revenue, including MaaS, is expected to surpass $100 billion, requiring an approximate 47% compound annual growth rate based on the current baseline.

This goal is quite aggressive, perhaps due to the company's strong financial performance.

According to the Tianyancha APP, in Q3 FY2026, Alibaba Cloud Intelligence Group's revenue reached RMB 43.284 billion, a 36% year-over-year increase. The MaaS platform's growth rate exceeded market expectations, with the Token consumption scale of the Bailian MaaS platform increasing sixfold over the past three months.

However, Alibaba is far from the only player in the Token economy. In this new resource competition centered on Tokens, major vendors are fiercely competing along different paths.

ByteDance has established a staggering advantage in Token usage through an 'application-driven, traffic-fed' model. According to Tan Dai, President of Volcano Engine, Doubao's large model has a daily Token usage exceeding 120 trillion, with the number of enterprises using over a trillion Tokens cumulatively increasing from 100 at the end of last year to 140.

Tencent's approach is more cautious, focusing on being a 'connector.' It does not pursue full-stack self-research but positions itself as an intermediary layer connecting model capabilities with rich application scenarios, deriving value through productization capabilities and ecological advantages.

Compared to these competitors, Alibaba has adopted the heaviest 'full-stack self-research' model.

Alibaba has achieved a full-stack layout in the AI field, covering chips (Pingtouge), computing power (Alibaba Cloud), models (Tongyi Qianwen), platforms (Bailian), and applications (Qianwen/Wukong), allowing for synergistic effects and building high technical barriers.

However, the flip side of full-stack investment is resource dispersion, with each area requiring substantial investment.

Alibaba has already announced plans to invest RMB 380 billion over the next three years in cloud and AI infrastructure, which may increase to RMB 480 billion.

However, according to Alibaba's Q3 FY2026 earnings report, the operating profit margin dropped to 4% from 15% year-over-year, operating cash flow decreased by 49% year-over-year, and free cash flow decreased by 71% year-over-year, with cumulative free cash flow for the first nine months at -RMB 29.3 billion.

The secondary capital market initially voted with its feet. Since January 22, Alibaba's stock price (NYSE:BABA) has been on a downward trend until April 7, when it began to climb as the market saw commercialization potential.

This raises the core contradiction faced by Alibaba AI's full-stack model: How to maintain massive investment while proving its ability to efficiently convert technological advantages into commercial returns?

Alibaba Cloud issued three consecutive price increase announcements in four days, with adjustments to the Bailian model service prices and DataWorks free quotas, reflecting the urgency of commercialization.

At the same time, Alibaba's AI centralization faces organizational-level challenges.

Can Alibaba break down 'departmental barriers' and truly reduce friction costs? Can organizational efficiency be sustained? After all, Alibaba AI now adopts horizontal management and lacks highly self-motivated technical talent like Lin Junyang. Can Alibaba AI, managed by KPIs, maintain its technological edge?

Additionally, the lack of high-frequency entry points similar to Douyin or WeChat requires higher costs for C-end application implementation.

Hence, despite Alibaba having accomplished its organizational transformation, the crux of the matter now lies in whether this 'centralized' organizational framework and 'full-stack' technical configuration can demonstrate their commercial efficacy amidst the intense market rivalry.

Conclusion:

The upheaval in March has not yet settled, and the decisive actions of April have already commenced.

From the silent exit of key personnel to the establishment of the ATH Business Group and the 'separation of powers' within the Technical Committee, Wu Yongming has, within a mere month, reshaped Alibaba AI from a fragmented entity into a unified force.

This 'Hundred Days' Reform, though lacking in overt drama, is nonetheless deserving of a place in the annals of organizational transformations within China's internet sector.

As each Token consumes computational resources and incurs costs in milliseconds, this battle is unceasing and irreversible.

Wu Yongming's decisive measures have addressed the entrenched issues. The pressing question now is: Can this colossus accelerate its pace despite the recent losses?

Disclaimer: This article is crafted based on legally disclosed corporate information and publicly accessible data, offering commentary thereon. However, the author makes no assurances regarding the completeness or timeliness of this information.

Additional Note: The stock market is fraught with risks, and caution is paramount when participating. This article does not serve as investment advice, and investment decisions should be made independently.