On March 16, the tech world was still buzzing with the aftershocks of Wu Yongming's company-wide memo.

The memo officially unveiled the launch of Alibaba's new core business unit, Alibaba Token Hub (ATH for short). This unit will operate on an equal footing with Alibaba's e-commerce and cloud intelligence divisions, with Wu Yongming taking the helm personally.

In his internal memo, Wu Yongming posited that we stand on the precipice of an AGI explosion, where the main actors in future digital work will be billions of AI Agents. The lifeblood of these Agents lies in the Tokens generated by models, making Tokens the primary conduit for human interaction with the digital realm. According to ATH's official definition, its core mission is succinctly captured as "creating Tokens, delivering Tokens, and applying Tokens."

This signifies Alibaba's intent to pivot entirely from "selling models" to "selling fuel." The underlying logic of this transformation, along with its implications for China's internet landscape, is far more intricate than a mere rebranding exercise.

Of course, it's also a high-stakes gamble: betting on the imminent arrival of the Agent era, the AI transformation of enterprise workflows, and the sustained leadership of the Tongyi Qianwen model in core capabilities. If successful, Alibaba will emerge as the "power grid" of the AI era. If it falters, this massive departmental integration could become another costly experiment in Alibaba's history.

I. From 'Feudal System' to 'Supply Chain'

To comprehend ATH, one must retrace Alibaba's AI strategy over the past few years to gain a clearer understanding of its rationale.

Alibaba's foray into AI began early, with the establishment of the DAMO Academy in 2017 to focus on foundational AI research. It subsequently launched the Tongyi series of large models, positioning itself among China's AI elite. Business units like DingTalk, Taobao, and 1688 also formed their own AI teams, attempting to weave AI capabilities into their operations.

However, the issue was glaring: these AI forces remained fragmented and operated in silos.

The Tongyi Lab concentrated on model development with little regard for commercialization. Alibaba Cloud's MaaS platform was tasked with model service output. The Qianwen APP focused on C-end personal assistants, with sluggish progress in integrating with the ecosystems of DingTalk and Taobao. DingTalk aimed to develop enterprise-grade AI applications but had to navigate through multiple business groups to access 1688's supply chain capabilities, often resulting in time-consuming and fruitless endeavors.

This "feudal fragmentation" might have allowed for multiple breakthroughs during the technological catch-up phase, but in a commercial competition demanding large-scale engineering delivery, it became a source of internal friction, failing to generate synergy.

From Alibaba's recent unification of the "Tongyi Qianwen" and "Qwen" brands, as well as the swift establishment of the ATH Business Group, it's evident that Alibaba's strategic direction is to no longer permit individual business units to operate independently and duplicate efforts in the AI realm. Instead, resources must be consolidated and operated with efficiency.

Specifically, ATH integrates five core sectors: The Tongyi Lab is responsible for foundational large models, akin to an "oil refinery," producing high-quality Tokens. The MaaS business line constructs the model service platform, functioning as "oil pipelines," efficiently and cost-effectively delivering Tokens to various scenarios. The Qianwen Business Division focuses on C-end personal AI assistants, the Wukong Business Division targets B-end enterprise-grade AI applications, and the AI Innovation Business Division explores new scenarios. These three act as "engines," converting Tokens into user and commercial value.

This "Tokenization" mindset represents Alibaba's third attempt at platformization in its history.

More than two decades ago, Alibaba platformized merchants, selling transaction opportunities. A decade ago, it platformized computing power, with Alibaba Cloud selling computing resources akin to utilities like water and electricity. Now, as model capabilities become homogeneous and computing power is no longer scarce, the ability to efficiently, securely, and cost-effectively generate and utilize Tokens has become the linchpin of enterprise competitiveness.

Wu Yongming aims to platformize "intelligence," selling Token consumption. In Alibaba's vision, future enterprises will not need to fret over underlying algorithms; they will simply pay for Tokens generated by each AI call, much like paying an electricity bill.

We believe Alibaba's transformation will likely compel other major domestic AI investors to expedite their AI organizational integrations. Previously, competition among these giants centered on model parameters and individual application experiences. Now, the focus will shift to resource integration capabilities and closed-loop capabilities from model to implementation, with organizational structural integration serving as the bedrock.

Alibaba is now completely liberating AI from its subordinate role under "the cloud," granting it equal status with cloud and e-commerce. This means Alibaba can no longer prioritize short-term gross margins on cloud resources; instead, it can undermine competitors' moats by lowering Token prices and leveraging scale effects.

After all, when Tokens become a universal basic commodity, whoever boasts the largest scale and highest distribution efficiency will wield pricing power.

II. A Historic Opportunity and an Uphill Battle

In management circles, there's a consensus: strategy (or structure) determines the possibility of success, while execution determines its reality. This holds true for Alibaba's transformation as well.

The first hurdle is whether Alibaba can truly dismantle its long-standing "departmental silos." Of course, these silos are not unique to Alibaba; they are a common affliction in almost all large companies.

Although ATH nominally consolidates the core frontlines related to AI, the organizational inertia within large companies is formidable. These teams possess vastly different genes, working styles, and evaluation standards, making it arduous to truly break down barriers and forge synergy.

The genes and core objectives of these teams diverge significantly: The Tongyi Lab emphasizes research, focusing on technological breakthroughs and metrics. The MaaS business line prioritizes ToB services, emphasizing stability, cost control, and commercialization. The Qianwen Business Division focuses on the C-end, concerning itself with user experience and traffic monetization. The Wukong Business Division delves into enterprise services, emphasizing industry adaptation and scalable implementation.

These teams previously had client-supplier relationships; now, belonging to the same business group, even with unified reporting lines, long-formed work habits and interest objectives are difficult to adjust swiftly. For instance, Tongyi's pursuit of technological breakthroughs may overlook costs, while MaaS's focus on commercialization requires cost control, leading to potential conflicts.

Similarly, the interest boundaries between ATH and Alibaba's existing business segments are difficult to delineate. ATH serves as Alibaba's AI core hub, but its relationship with existing business lines like Alibaba Cloud, DingTalk, Taobao, and Tmall—whether it empowers or controls them—remains nebulous. For example, Alibaba Cloud was previously the core outlet for Alibaba's AI external output; now, with the MaaS business line under ATH, how will Alibaba Cloud and ATH divide labor and distribute benefits?

Alibaba has attempted organizational integrations multiple times in the past, such as the "1+6+N" organizational transformation in 2018, aiming to break down departmental silos and achieve resource synergy. However, the results were lackluster, with many departments continuing to operate independently and even new internal frictions emerging.

If the interest boundaries are not clearly defined and true synergy is not achieved among teams in this ATH integration, then the establishment of ATH loses its purpose.

Beyond dismantling departmental silos, Alibaba's "All in" bet on the AI Agent era is predicated on a premise: the Tongyi Qianwen model must maintain absolute leadership in core capabilities.

ATH's intricate "Token supply chain" is built on the performance of its in-house models. If the Tokens produced by the Tongyi Lab lag behind competitors in quality, long-text processing capabilities, or multimodal understanding by a generational gap, then the entire ATH supply chain may face risks of "overcapacity" or "technological obsolescence."

On the eve of ATH's establishment, the departure of Lin Junyang, the soul figure behind Tongyi Qianwen, reflected, to some extent, the friction between "scientist leadership" and "industrialized management." If Alibaba weakens its exploration of foundational original technologies in pursuit of extreme engineering and Token consumption, then once open-source communities or agile teams like Kimi and DeepSeek achieve algorithmic breakthroughs, the "create-deliver-apply" closed loop built by ATH will lose its core support.

This is akin to investing billions in an oil refinery, only to find that engine technology has skipped fossil fuels and entered the nuclear era. No matter how extensively the refinery's pipelines are laid or how efficient they are, they would only be managing a pile of scrap iron.

For Alibaba, the establishment of ATH is just the first step. The real test lies in how to streamline internal interest relations, break down departmental silos, and truly forge synergy among teams. It must continue to ramp up R&D investment to maintain the leadership of the Tongyi model and withstand industry competition. It also needs to rapidly promote AI application implementation, convert the value of Tokens into tangible commercial gains, and make ATH truly Alibaba's core growth engine for the future.