Alibaba's Token Foundry Reorganization Comes With Profit Strain

Alibaba has reorganized its artificial-intelligence teams again, creating a Token Foundry business unit that will report directly to chief executive Eddie Wu. The move places more of the company's AI execution under central command, but it also arrives at a moment when Alibaba's investment burden is visibly damaging profitability.

The new unit combines the Tongyi large-model business and Future Life Lab. Zhou Jingren will become Alibaba's chief scientist and lead a new AI future research institute focused on frontier AI work. Zheng Bo will bring projects including Happy Horse and Happy Oyster into Token Foundry. The structure is designed to put model development, product exploration and future AI research closer to group-level strategy.

Alibaba has reasons to move quickly. Its Qwen models have become among China's most important open and commercial model families, with the company describing Qwen-3.7 as highly competitive in coding and developer use. Cloud demand tied to AI has also become one of Alibaba's stronger growth stories, with management saying AI-related revenue has become a material part of external cloud commercialization.

Yet the reorganization should not be treated as a clean victory. Alibaba's most recent quarterly numbers showed the cost of the AI and user-experience push. Revenue for the March quarter was 243.38 billion yuan, up 3 percent. On a comparable basis excluding disposed businesses, growth was stronger. But operating performance deteriorated sharply. The company reported an operating loss of 848 million yuan, compared with operating profit of 28.47 billion yuan a year earlier. Adjusted EBITA fell 84 percent to 5.10 billion yuan, and non-GAAP net income dropped almost completely from the prior-year level.

Those figures make the AI story more difficult. Alibaba can argue that heavy spending is necessary to defend its position in cloud, models, instant retail and user experience. Investors can accept investment cycles when the payback path is visible. The problem is that AI infrastructure and product development can absorb enormous capital before producing stable margins, especially when competitors are also subsidizing model access, cloud discounts and developer adoption.

A direct reporting line to Wu may improve coordination, but it can also reveal that previous structures were not decisive enough. Alibaba has already moved Tongyi from a lab structure into a business unit and adjusted senior technology responsibilities earlier this year. Frequent reorganization can be a sign of urgency; it can also become a sign that the company is still searching for the right operating model.

The word 'Token' in the new unit's name is revealing. The economics of AI are increasingly measured not only in model rankings, but in token volume, inference cost, developer usage and conversion into paying enterprise workloads. A model can be impressive and still fail to generate enough high-margin usage to justify compute spending. Alibaba now has to prove that Qwen can become an economic engine, not only a technical asset.

The cloud business is the most logical route to monetization. If Qwen drives enterprise cloud usage, custom model deployment, inference workloads and data-service demand, Alibaba can turn AI leadership into infrastructure revenue. But cloud customers are price sensitive, and China's model market remains highly competitive. Open models can build developer goodwill while also making it harder to charge premium prices.

The reorganization also raises execution risk across consumer products. Projects such as Happy Horse and Happy Oyster suggest Alibaba is still looking for AI-native applications beyond infrastructure. That search is necessary, but it is expensive and uncertain. Consumer AI products can attract attention quickly and lose it just as fast if the use case is not habitual.

For Alibaba's broader business, the danger is that AI spending collides with other strategic drains. The company is investing in instant retail, defending e-commerce engagement and improving user experience. Each priority may be rational on its own, but together they weaken near-term earnings and make capital allocation harder to explain.

The negative reading is that Alibaba is compressing several fights into the same period: model competition, cloud monetization, consumer-app experimentation, retail defense and organizational reform. A single new unit cannot simplify all of that.

The test for Token Foundry is therefore commercial, not ceremonial. Alibaba needs fewer internal titles and more evidence that AI use is raising cloud margins, locking in enterprise customers and creating products that users pay for. Until then, the reorganization may be remembered less as a breakthrough than as another sign of how costly the AI race has become.