China's Token Invocations Reach 21,100 Trillion in 2025

Authoritative Report Released: China's AI Invocation Volume Sees Explosive Growth

At the 9th Digital China Summit, the National Data Resources Survey Report (2025) was officially released. The report disclosed a striking set of figures — in 2025, China's cumulative annual token invocations reached approximately 21,100 trillion, with daily average invocations soaring from over 1 trillion at the beginning of the year to 100 trillion by year-end, tracing a classic exponential growth curve. These numbers not only showcase the robust vitality of China's artificial intelligence industry but also signal that data resources are empowering AI innovation at an unprecedented scale.

Token Invocations: The Core Metric for Measuring the AI Industry

A token is the smallest unit of information created when AI large language models break down text, images, audio, and other data for efficient processing. Every user conversation with a large model, every piece of AI-generated content, and every inference made by an intelligent system generates a massive volume of token invocations. As such, token invocation volume has become a key indicator of AI industry activity and application penetration depth for any country or region.

According to the report, the national daily average token invocation volume grew nearly 100-fold over the course of 2025. At the start of the year, daily invocations had just crossed the trillion mark; by year-end, the figure had climbed to 100 trillion — a growth rate far exceeding industry expectations. The full-year cumulative total of 21,100 trillion token invocations reflects the fact that large model technology has comprehensively transitioned from the laboratory stage to the stage of scaled deployment.

Three Major Drivers Behind the Exponential Growth

First, the broad rollout of large model application scenarios. In 2025, AI large models accelerated their adoption across office collaboration, content creation, intelligent customer service, educational tutoring, medical consultations, code development, and more. From enterprise-level applications to consumer-facing products, the frequency and depth of large model usage increased significantly, driving sustained growth in token consumption.

Second, the rapid proliferation of multimodal large models. Compared to text-only models, multimodal large models generate several times — or even tens of times — more token consumption when processing images, video, and audio data. In 2025, multiple domestic companies launched mature multimodal products, and the explosion of use cases such as video generation and image-text comprehension directly elevated the overall scale of token invocations.

Third, the continuous improvement of AI infrastructure. Enhanced computing power supply and dramatic reductions in inference costs have significantly lowered the barrier to using large models. According to industry observations, inference costs for mainstream domestic large models dropped by more than 80% year-over-year in 2025. The resulting "token price war" objectively accelerated the mass adoption of AI applications, indirectly boosting the total volume of token invocations.

Data Resources Empowering AI: From Quantitative to Qualitative Change

The report specifically emphasized that China's data resources are empowering AI innovation with "greater intensity." Behind this statement lies the ongoing deepening of market-oriented reforms for data as a factor of production. As mechanisms for open sharing of public data improve and data trading infrastructure continues to develop, the supply capacity for high-quality training data has been significantly enhanced, providing solid "fuel" for the iterative upgrading of large models.

Notably, the exponential growth in token invocations also places higher demands on data security, privacy protection, and computing resource allocation. How to ensure compliant data usage and green, sustainable development while driving the expansion of AI applications will be a critical challenge in the next phase.

Outlook: 2026 May See an Even Larger-Scale Explosion

Based on the current growth trajectory, China's token invocation volume is expected to multiply several times over in 2026 compared to 2025 levels. As new paradigms such as AI Agents and embodied intelligence gradually mature, large model invocation scenarios will extend further into the physical world, potentially triggering yet another order-of-magnitude leap in token invocations.

From trillions to hundreds of trillions, this steep growth curve not only records the speed at which China's AI industry is advancing but also heralds the accelerating arrival of a new economic era driven by data and intelligence.