Alibaba recently announced the creation of a new "Token Foundry" unit, consolidating its key AI development teams under one roof.
This move didn't happen in a vacuum; it's a direct response to a dramatic surge in demand for its AI services. The numbers speak for themselves: Alibaba reported that AI token consumption on its platform grew sixfold in just three months, and average daily token revenue later surged fifteenfold. AI products now make up about 30% of Alibaba Cloud's revenue, which itself is growing at a rapid clip. This reorganization is essentially Alibaba's way of building a faster, more efficient engine to meet this booming customer demand.
The creation of the Token Foundry is also the final step in a series of organizational changes over the past few months. First, in March, the company formed the "Alibaba Token Hub" as an initial effort to bring its AI teams together. Second, this was followed by the creation of a CEO-led technology committee in April to centralize decision-making. These earlier steps paved the way for the Token Foundry, which formalizes the consolidation and aims to create a more agile and focused R&D pipeline for its core Qwen models and enterprise AI agents.
Finally, there's a crucial hardware component to this story. Alibaba is navigating a complex technological landscape, balancing the use of high-end overseas GPUs, which are subject to U.S. export restrictions, with its own homegrown AI chips like the Zhenwu M890. By unifying its model development teams, the company can better coordinate how it trains and deploys its AI models across this mixed-hardware environment, ensuring it gets the most out of its valuable computing resources.
Despite the strategic logic, the initial market reaction was mild. Alibaba's stock price didn't see a boost, partly because its valuation is already high compared to its own history. This suggests investors are waiting for proof. The creation of the Token Foundry sets the stage, but now Alibaba must deliver sustained growth in AI revenue to justify the market's high expectations.
- Token: In AI, a token is a piece of text, like a word or part of a word, that a language model processes. Token consumption is a key metric for measuring AI service usage.
- Foundation Model: A large, powerful AI model (like Alibaba's Qwen) trained on a vast amount of data, which can be adapted for various specific tasks.
- Compute Cluster: A group of many powerful computers (often with GPUs) networked together to perform intensive tasks, such as training a large AI model.