ByteDance’s cloud arm, Volcengine, has reached a significant milestone in the global AI race, announcing that its Doubao large language model (LLM) has surpassed 120 trillion daily token usage as of March 2026. This represents a staggering thousand-fold increase since its launch in May 2024, signaling a rapid maturation of China’s 'Model-as-a-Service' (MaaS) ecosystem where tokens have effectively become the new currency of industrial productivity.
During the 'AI Innovation Exhibition' in Wuhan, Volcengine President Tan Dai articulated a shift in the corporate strategy, moving beyond mere model development to the creation of a 'Token Economy.' The number of enterprises consuming over a trillion tokens has jumped by 40% in just a few months, reflecting a trend where AI is no longer a peripheral experiment but a core operational engine for Chinese industry. This surge is being driven by the integration of multi-modal capabilities and the proliferation of desktop intelligent agents.
To solidify its position against heavyweights like Alibaba and Tencent, Volcengine also launched the public enterprise beta of Seedance 2.0, its latest video generation model. Addressing the critical bottlenecks of copyright and 'deepfake' concerns, the API includes built-in detection and defense mechanisms for intellectual property and portrait security. This move aims to transition AI video from a consumer novelty into a reliable tool for industrial-scale content production.
The competitive landscape in China is tightening as Alibaba recently elevated its MaaS business to a group-level strategic track under CEO Eddie Wu, while Tencent Cloud revamped its own 'TokenHub' to unify various models including its internal Hunyuan and the popular DeepSeek. Volcengine is countering this by focusing on 'Agentic' efficiency, arguing that the true cost to enterprises is not the unit price of a token, but the 'wasted tokens' consumed during inefficient AI reasoning and trial-and-error processes.
By launching the ClawHub China mirror station in partnership with the OpenClaw project, Volcengine is also attempting to build an open ecosystem for AI agents. Tan Dai envisions a dual-track development for corporate AI: 'Agile Agents' for individual productivity and 'Stable Agents' for institutional workflow scaling. This strategy suggests that the next phase of the AI war in China will be won not just by the smartest model, but by the platform that offers the most efficient and secure integration into existing business logic.