ByteDance Unit Reports 50 Trillion Daily AI Tokens, Extending Lead in China Market
ByteDance’s cloud division, Volcano Engine, reported that daily AI token usage for its rigorous Doubao models has reached 50 trillion, a figure that underscores the company’s aggressive expansion in the generative AI sector and firmly establishes its dominance in the domestic market.
The reported volume represents a 417-fold increase since the model's launch and a more than 10-fold jump compared to December 2024. This daily operational scale now surpasses comparable figures released by Alphabet Inc.’s Google in October, positioning the Chinese tech giant as a top global player in AI inference demand alongside U.S. counterparts.
Volcano Engine President Tan Dai stated at the company's FORCE conference that its share of China's Model-as-a-Service (MaaS) market has expanded to 49.2% in 2025, up from 46.4% the previous year. The executive noted that the sheer volume implies that half of all AI tokens generated on China's public clouds are now processed through Volcano Engine’s infrastructure.
This growth is underpinned by a strategic pivot toward a "Token Economy," where infrastructure is explicitly retooled to support model inference and agent-based workflows rather than traditional compute services. To sustain this demand, the company introduced the updated Doubao 1.8 model and the Seedance 1.5 Pro video model, aiming to capture the structural shift toward multimodal and agent-driven AI consumption.
Surging Demand and Market Share
The 50 trillion daily token figure serves as a key indicator of actual AI adoption capabilities, distinct from theoretical computing power. According to International Data Corporation (IDC) reports cited by the company, Volcano Engine has secured the premier spot for public cloud large model service calls in China.
The trajectory of this growth highlights a rapid commoditization of AI outputs. Wu Di, head of intelligent algorithms at Volcano Engine, projected that domestic token consumption could expand more than 100-fold by 2030. Under this forecast, the primary metric for valuing corporate intelligence capabilities will shift from the quantity of GPUs owned to the total volume of tokens consumed, reflecting a transition to usage-based value assessment.
Redefining Cloud Infrastructure
To accommodate this volume, Volcano Engine is restructuring its cloud architecture, moving away from the traditional IaaS/PaaS/SaaS hierarchy toward a model-centric "AI Cloud Native" framework. The company argues that the legacy approach of renting GPUs and databases is becoming obsolete for modern AI developers.
New offerings include "Inference Foundry," allowing enterprises to host models with pricing based strictly on token consumption rather than server rental. Additionally, the updated Agent Development Kit (ADK) introduces a multi-session shared runtime model, contrasting with the distinct runtime architecture used by competitors like AWS. This design choice implies a focus on cost reduction and efficiency, leveraging ByteDance’s internal experience in managing massive, high-concurrency requests for apps like Douyin.
The Token Economy Strategy
The company’s "Token Economics" thesis posits that tokens are becoming the fundamental unit of digital intelligence. While current consumption is driven largely by inference replacing training, future growth is expected to stem from multimodal interactions and autonomous agents.
Tan Dai suggested that the market is evolving from a raw material phase—where tokens are viewed as IT budget items—to an agent-based phase akin to Business Process Outsourcing (BPO). By 2027 or 2028, the company anticipates a decline in text-heavy interactions in favor of visual and agent-driven signals, fundamentally altering the operating logic of the digital layer. Real-world data reveals that over 100 enterprise clients have already exceeded a cumulative usage of 1 trillion tokens on the platform.
Vertical Integration and Product Strategy
ByteDance's approach mirrors the vertical integration strategy seen at Google, distinguishing it from the partnership model of OpenAI and Microsoft Corp. or the open-source emphasis of Alibaba. The Doubao model functions as a unified "All-in-One" foundation, integrating text, code, audio, and video capabilities to minimize client selection costs.
This integration allows for immediate commercial validation of research. Market feedback gathered by Volcano Engine directly influences the research and development decisions of the Seed model team. Features deemed high-value for business metrics—such as built-in tools and draft previews for video generation—are prioritized over technical specifications with low practical utility, such as excessively long context windows.