As competition in China's artificial intelligence (AI) industry shifts quickly, Alibaba has overhauled its AI organization. With the recent boom in China for the AI agent development framework "OpenClaw" and the rapid spread of competition in related services, Alibaba is seen as expanding its AI strategy from a focus on large models to a contest centered on AI agent-based applications and platforms.

On the 17th, according to China Business News, Alibaba recently created a new "Alibaba Token Hub" business team. The new unit integrates AI model development, model services, and application services into a single operating framework. Basic model research, the model service unit for corporations, and AI application divisions have been consolidated under this organization. Chief Executive Officer Wu Yongming is directly overseeing it.

OpenClaw logo. /Courtesy of Reuters

◇ The OpenClaw craze shaking China is the backdrop

The reorganization coincides with recent changes in China's AI industry landscape. Over the past few years, China's big tech companies have focused on competing over large language model (LLM) performance. But as major companies have succeeded in securing their own models and aggressively rolling out services, narrowing performance gaps, the focus of competition has shifted to AI application services.

In particular, the rapid spread and rising popularity of an AI agent development framework called "OpenClaw" in China's developer community has significantly changed the industry mood. OpenClaw is a tool that implements an agent architecture in which AI plans and executes tasks on its own, different from traditional chatbots. While chatbots generate answers to questions, AI agents like OpenClaw receive goals from users and directly produce outcomes. They go beyond simply explaining information to autonomously handling real work such as data collection, analysis, and program execution. It is a progression from the "answering stage" to the "doing stage."

OpenClaw began gaining major traction in China's tech sector in early March. On social media, installation methods, model combinations, and usage tips for OpenClaw spread rapidly, and paid services offering installation help emerged at high prices. When Tencent said it would support free installation, around 1,000 developers and others seeking setup gathered in front of its Shenzhen headquarters. As OpenClaw's popularity turned into a frenzy, major local corporations including Alibaba, ByteDance Ltd., Baidu, SenseTime, MiniMax, and Zhipu unveiled related services in quick succession, intensifying the competition.

Graphic by Son Min-gyun

◇ Alibaba shifts AI strategy, betting big on "tokens"

Amid these changes, Alibaba's reorganization is seen as an attempt to move the center of its AI strategy from "basic model development" to "AI agent competition." While Alibaba has posted some results in the model development race, it has been viewed as having a weaker presence in the generative AI app market—epitomized by chatbot services—compared with ByteDance Ltd.'s "Doubao." Separate from technical prowess, the company has been seen as lagging in securing real user touchpoints.

Aware of these limits, Alibaba appears set to focus less on the model itself and more on controlling the structures in which AI is actually used. As AI agents spread, they are expected to handle more tasks than human engineers, and in that process, usage of "tokens," the computational units of AI, is projected to surge. Experts say the core of future AI competition will be how many tokens companies can create, distribute, and drive consumption for.

In response, Alibaba set "creation, supply, and utilization of tokens" as the new unit's core goals. It plans to generate tokens through multimodal model development, supply them via a large-model service platform for corporations, and expand token usage through AI application services for businesses and individuals. In an internal letter, CEO Wu said, "We are currently on the eve of an AGI explosion," adding, "In the future, billions of AI agents will perform digital tasks, and as they operate based on tokens generated by models, they will become the key medium connecting humans and the digital world."

※ This article has been translated by AI. Share your feedback here.