- IDC expects China's token consumption to surge 20-fold in 2026 compared to 2025, with total annual consumption poised to reach 40 quadrillion.
- China's daily average token usage already topped 140 trillion in March this year.
China's token economy is entering an explosive phase, shifting from conceptual pilots to large-scale applications as artificial intelligence (AI) technology accelerates its rollout across domestic companies.
China's token consumption is expected to surge 20-fold in 2026 from 2025 levels, with total annual usage poised to hit 40 quadrillion, according to a report released Thursday by market research firm IDC.
This accelerated growth trajectory is primarily driven by the growing maturity of multimodal large models and the large-scale deployment of various AI agent applications.
The forecast echoes robust data released by Chinese authorities, further confirming the rapid expansion trajectory of the domestic AI industry.
China's daily average AI token usage topped 140 trillion in March, surging more than 1,000 times from the level at the beginning of 2024, the National Data Administration said at a media briefing in March.
As the number of active AI agents continues to climb, a new commercial value system centered around token usage, distribution, and settlement is rapidly evolving.
Despite the exponential expansion of the overall market size, massive discrepancies in token consumption levels remain across different vertical application scenarios.
Currently, the public cloud Model as a Service (MaaS) market is highly concentrated among tech giants like Volcano Engine and Alibaba Cloud, while the private deployment market favored by government and enterprise clients remains far more fragmented.
Measured by usage volume, Volcano Engine captured nearly half of the market share in 2025, followed by Alibaba Cloud, Baidu AI Cloud, SiliconFlow, and China Mobile Cloud, according to IDC.
As the market gradually matures, the competitive focus among MaaS providers is shifting entirely from simple price wars to a comprehensive contest of price, performance, and toolchains.
Notably, behind this booming growth, enterprises still face multiple practical constraints and medium-to-long-term risks during the implementation of large models.
When deploying large models, companies currently prioritize model performance, security compliance, and response quality over cost-effectiveness, IDC noted.
If severe computing power bottlenecks emerge or compliance policies change in the future, the actual growth rate could fall significantly short of the high-growth expectations, according to IDC.
