Taking advantage of the aftershocks of Lin Junyang's (林俊旸) resignation, Wu Yongming (吴泳铭) has dismantled the old load-bearing wall of Alibaba AI (阿里AI)

Rebuild from the top down

It has been reported by TMTPost that Alibaba (阿里巴巴) CEO Wu Yongming (吴泳铭) used the momentum after Lin Junyang’s (林俊旸) departure to push through a sweeping realignment of the group’s AI organization. In two corporate letters over three weeks, Wu named clear owners for model development, cloud infrastructure and inference execution — Zhou Jingren (周靖人) to run the Qwen/通义 large‑model business, Li Feifei (李飞飞) as Alibaba Cloud CTO to stabilize AI cloud infrastructure, and Wu Zeming (吴泽明) to take group CTO duties plus the AI inference platform — and formed a technical committee chaired by Wu. The message: no more fuzzy metrics, no more parallel AI wheels in separate business units.

Token as currency, ATH as central bank

At the centre of the redesign is ATH (Alibaba Token Hub, 阿里Token Hub) and a plan to make "Token" the unified unit of measurement and settlement across the group’s AI stack. Token is being elevated from a technical billing unit to an internal currency that can quantify model inference, API calls and agent executions across commerce, logistics and cloud. Why does that matter? It replaces ambiguous budget swaps and executive coordination with a programmable, auditable ledger of AI consumption — ATH becomes an internal clearinghouse that can price, meter and settle cross‑business usage.

From race horses to concentrated firepower — and the risks

The move marks a pivot away from the old "赛马" (race‑to‑win) decentralised model toward centralized command: concentrate compute, data and engineering resources to scale models and amortize ongoing inference costs. It is a bet on infrastructure over isolated model glory. But risks remain. Centralisation can create single‑point failures, and execution depends on stitching together very different KPIs (developer retention, DAU, GMV) into a Token economy that businesses accept. It has been reported that Wu previously set an ambitious target of roughly $100 billion in annual AI and cloud revenue within five years; achieving that will require both technical depth and commercial buy‑in.

Geopolitics and the wider context

For Western readers: this restructuring happens against a backdrop of U.S. export controls on advanced chips and intensifying Sino‑U.S. technology rivalry, which have pushed Chinese tech groups to optimize software, system integration and domestic compute architectures rather than rely on imported hardware alone. Reportedly, Alibaba’s aim is to lock in an internal flywheel — models, cloud and inference billed by Token — so that AI capacity becomes a monetizable, cross‑BU platform rather than a set of siloed experiments. The architecture is clear; the hard part will be turning Token thinking into predictable revenue while managing regulatory, supply‑chain and operational friction.