Alibaba launches another new business division — what signal does this send?
The tech circle in June has been anything but calm.
A couple of days ago, Alibaba unveiled its third wave of AI organizational adjustment plans since 2026. However, this time it's not just a minor personnel change. Instead, it directly merged two major AI teams, the Tongyi Large Model Division and the Future Life Laboratory, to form a brand - new Token Foundry Division, led personally by Group CEO Wu Yongming.
Meanwhile, Zhou Jingren, the founding figure of Alibaba's Qianwen system, officially assumed the position of the group's Chief Scientist and took the lead in establishing the Alibaba AI Future Research Institute. Zheng Bo, the former head of the Future Life Laboratory, led star product teams such as HappyHorse and HappyOyster to be integrated into the new division as a whole.
As soon as the news broke, the entire industry started discussing the same question: What's Alibaba up to this time?
Just over three months ago, Alibaba established the AlibabaTokenHub (ATH) Business Group, led by Wu Yongming. It brought together AI - related units such as the Tongyi Laboratory, the MaaS business line, the Qianwen Division, the Wukong Division, and the AI Innovation Division under one organizational framework. Two months ago, he also formed the Group Technology Committee and served as its leader.
The signals released in the previous two adjustments were clear: Reduce the collaboration costs between different business lines and form a synergy for AI at the organizational level.
Just a few months later, the Token Foundry Division was established, which means that Alibaba's AI strategy has shifted from the "resource integration" stage to the "accelerated implementation" stage.
The Ambition of the "Token Factory"
The name "Token Foundry" is quite interesting. The word "Foundry" literally means a foundry or a metal factory. When combined with "Token", it becomes even more meaningful. Alibaba aims to transform itself into a "Token factory" and become a base - level supplier in the AI era.
This is in line with Alibaba's strategic orientation. In March this year, when Alibaba established the ATH Business Group, a clear goal logic was set internally: "Create Tokens, deliver Tokens, and apply Tokens." The establishment of the Token Foundry Division now seems to be an extension and deepening of this logic.
From the perspective of the organizational structure change, before the merger, the Tongyi Large Model Division was responsible for the research and development of the underlying models, while the Future Life Laboratory focused on AI scenario exploration. Although both were working towards the same major goal, they were on different reporting lines, and it was difficult to avoid collaboration costs.
The Future Life Laboratory was previously affiliated with Taotian Group and was later transferred to the newly established ATH Business Group. It had been exploring AI applications. The Tongyi Large Model Division was also doing similar work, such as the research and development of the video - generation model, and released the latest version, Wanxiang 2.7, this year. Each side was working independently, resulting in resource duplication and internal competition - like inefficiencies. After the merger, theoretically, resources can be concentrated on the "most critical battlefield" to avoid the situation of working in isolation.
An industry insider told the media that the arrival of the Agent era has brought obvious organizational changes: When developing chatbots, the model team can be less involved in the business; but when developing Agents that can autonomously execute workflows, the model team must understand the business logic, data flow, and decision - making chain.
There is also a detail that is easily overlooked but worthy of attention: Zheng Bo led projects such as HappyHorse and HappyOyster to join the Token Foundry Division.
The name HappyHorse became quite popular in the AI circle in April this year. It anonymously participated in the global authoritative AI blind - testing platform ArtificialAnalysis and topped both the text - to - video and image - to - video tracks, attracting a lot of industry discussions and attention with its performance.
The fact that a product from the Future Life Laboratory could suddenly emerge and achieve world - class results shows that there are "good products" within Alibaba. The question is whether such good products can be continuously produced and systematically introduced to the market. The joining of Zheng Bo's team indicates that Alibaba wants to integrate this ability into a larger framework.
Of course, whether the organizational adjustment can truly solve the collaboration problem remains to be seen. The merger is just the first step, and the real tests lie in the subsequent cultural integration, process streamlining, and goal alignment. It's easy to combine two departments, but difficult to make them truly have a chemical reaction.
So far, Alibaba's AI organizational structure has become very clear: The ATH Business Group serves as the top - level structure, coordinating all AI businesses; the Token Foundry Division is responsible for model research and development and commercial implementation; the AI Future Research Institute focuses on cutting - edge technology exploration; the MaaS business line builds a model service platform; the Qianwen Division creates a C - end personal AI assistant; and the Wukong Division builds a B - end AI - native work platform.
This four - layer structure of "research institute - basic model - service platform - application product" not only ensures the long - term technological innovation ability but also takes into account the short - term commercialization needs. Wu Yongming has built an efficiently operating organizational machine for Alibaba's AI in just three months.
Zhou Jingren's Transition: What's Alibaba Thinking?
In this organizational adjustment, another role change worth pondering is Zhou Jingren.
This key figure in Alibaba's Tongyi large - model team once served as the Chief Scientist of Alibaba Cloud and was in charge of data - intelligent businesses such as search, recommendation, and advertising for Alibaba Cloud, Taobao, and Alipay. At the end of 2022, he began to serve as the CTO of Alibaba Cloud Intelligence, concurrently as the deputy dean of Alibaba's DAMO Academy and the head of the Tongyi Laboratory. He built the Tongyi large - model team from scratch, promoted the Qwen series of models from 0 to 1, and finally made them rank among the world's first - tier models.
The newly released Qwen - 3.7 model has achieved the world's top - three and China's top - one coding ability, and has been widely recognized in the developer community and among industry customers. In 2025, Zhou Jingren became a partner of Alibaba, being the first CTO - level executive with a pure technical background to enter the highest decision - making level.
In his new role as the group's Chief Scientist, Zhou Jingren will no longer be responsible for specific business management but will fully immerse himself in the research of cutting - edge AI technologies.
The title of Chief Scientist is the highest academic title in Alibaba's technical system. Zhou Jingren joined the Alibaba partnership last year. In less than a year, he obtained the highest academic title. This promotion pace is worth pondering. Obviously, Alibaba hopes that he can "travel light" and be freed from specific business management to focus on more long - term technological propositions.
This kind of arrangement is not uncommon in the industry. OpenAI has a dedicated super - alignment team, and Anthropic also has its own cutting - edge research department. When large - model technology develops to a certain stage, someone needs to be freed from the daily product iteration and commercialization pressure to focus on more long - term technological breakthroughs.
This organizational design actually reflects Alibaba's layout idea: One hand focuses on the present, and the other on the long - term. The Token Foundry Division is responsible for productization and commercial implementation, with the CEO personally overseeing the progress; the AI Future Research Institute is responsible for cutting - edge exploration, led by the Chief Scientist for basic research.
The two lines should advance simultaneously, ensuring that the business does not fall behind in technological iteration and that long - term competitiveness is not lost due to excessive pursuit of short - term interests.
Zhou Jingren once had a clear statement about the development trend of large models: "Large models are undergoing a core paradigm shift, from aligning with human preferences to aligning with task goals. In the past, we pursued models that 'talk well'; now we require models that 'can do things'."
This shift from "good - looking indicators" to "reliable performance" may better represent Alibaba's real attitude.
Currently, there is not much official information about the future research direction of the AI Future Research Institute. However, based on Zhou Jingren's previous technological judgments, the ability of autonomous planning, continuous iteration, and cross - tool collaboration, that is, the so - called Agent ability, may be an important area of focus.
Zhou Jingren previously clearly stated that with the leap in the capabilities of the Qwen - 3.7 series of models, Alibaba is making the model truly become the intelligent core of Agents.
A "Organizational Competition" is Unfolding Among Big Tech Companies
If we look at the entire industry, we'll find that Alibaba's adjustments are not an isolated case. In the past two years, almost all top - tier AI companies have been undergoing similar organizational restructurings, and the steps are surprisingly similar.
Previously, Google merged its nearly - decade - old Brain team and DeepMind into GoogleDeepMind, led by Demis Hassabis, who reports directly to CEO Sundar Pichai. At the beginning of last year, Google went a step further and integrated all AI engineering teams scattered across various product lines under DeepMind, completely completing the centralization of its AI organization.
On Microsoft's side, in 2026, it reorganized the Copilot team and set up an executive vice - president position reporting directly to CEO Satya Nadella, with the goal of shortening the decision - making chain. Meta restructured its AI organization four times within half a year in 2025, with the core direction of bridging the gap between the FAIR Laboratory and the product AI team. Amazon merged its AGI team, self - developed chip team, and quantum computing team into a unified organization between 2025 and 2026, integrating the entire chain from infrastructure to model research and development.
Behind these actions, we can identify three common patterns: First, AI is moving from "independent operation in the laboratory" to "deep integration with the business"; second, the reporting relationship has been upgraded from the vice - president level to direct management by the CEO or president; third, models, infrastructure, and products are no longer under different command systems but are integrated into the same combat unit.
So, looking at Alibaba's recent move, the establishment of the Token Foundry Division is essentially in line with this global trend. The difference is that Alibaba has completed the intensive promotion from ATH to Token Foundry in just a few months, which is the fastest pace among domestic Internet giants.
In reality, any organizational adjustment ultimately boils down to business. Behind Alibaba's bold integration of AI business, there is a key time point: In May, Alibaba Group Chairman Joe Tsai and CEO Wu Yongming jointly issued a letter to shareholders, announcing that Alibaba's AI business has passed the initial investment stage and officially entered the commercial return period.
In other words, AI within Alibaba needs to gradually take on the important task of "increasing revenue". The financial report shows that in the fourth quarter of fiscal year 2026, the external commercial revenue of Alibaba Cloud accelerated to 40% growth, and the revenue of AI - related products has achieved triple - digit growth for the eleventh consecutive quarter.
Wu Yongming revealed a more specific figure at the earnings conference: The annual recurring revenue (ARR) of AI models and application services, including the Bailian MaaS platform, will exceed 10 billion yuan in the June quarter and 30 billion yuan by the end of the year.
He also mentioned that the API demand for the Bailian platform has increased by more than 10 times in the past six months. "There are hardly any empty cards in our servers, and there are still many customers in the queue." In the highly competitive AI track, this state of supply falling short of demand is the greatest competitive advantage.
Alibaba's newly released Qwen - 3.7 model ranks fifth globally and first among domestic models in the large - model intelligence ranking of the third - party institution ArtificialAnalysis, and has been widely recognized in the developer community and among industry customers.
Of course, we also need to see that although the commercialization data is impressive, the competition pressure is not small compared with peers. In the MaaS (Model as a Service) field, for example, bytedance's Volcengine has raised its MaaS target for this year from 10 billion to 15 billion. Liu Weiguang, a senior vice - president of Alibaba Cloud Intelligence Group, also stated that he has issued a military order to the sales team: By the end of the year, the MaaS revenue of each customer should account for no less than 20%.
The competition in the MaaS track is becoming extremely fierce, and the establishment of the Token Foundry Division is, to some extent, a new move by Alibaba at the MaaS table.
Looking at the long - term, Alibaba's actions in the AI field this year can be described as "high - frequency". The intensive pace reflects the reality that the window period for the AI industry is rapidly narrowing.
In 2026, the enthusiasm for the parameter competition gradually cooled down, and instead, there is a comprehensive competition in engineering capabilities, commercialization levels, and ecosystem integrity. The daily active users (DAU) of ByteDance's Doubao have exceeded 200 million, and Tencent's Hunyuan Hy3Preview is being widely applied in products such as CodeBuddy, WorkBuddy, and Yuanbao.
It's worth noting that the statement in Alibaba's letter to shareholders this year that "the AI business has passed the initial investment stage and officially entered the commercial return period" is almost at the same time as ByteDance's announcement that Doubao will start paid subscriptions. The fact that leading players are entering the commercial fast - lane simultaneously means that the competition relying solely on technology and product innovation is turning into a three - line race of technology, products, and commercialization.
The establishment of the Token Foundry Division can be seen as a strategic move by Alibaba at this critical juncture: It has both a long - term layout for the future and a timely response to the immediate situation.
This article is from the WeChat official account "New Vision" (ID: xinmouls), written by Li Xiaodong, and is published by 36Kr with authorization.