China’s CAICT says public‑cloud large‑model usage jumped 16× as AI core industry heads past 1.2 trillion yuan

Summary

China’s Academy of Information and Communications Technology (CAICT) estimates that in 2025, calls to public‑cloud large language models exceeded 20 quadrillion tokens, up roughly 16× year‑on‑year, with ByteDance’s Volcano Engine alone seeing a 250× surge in daily token usage since April 2024. ([companies.caixin.com](https://companies.caixin.com/2025-12-15/102393332.html)) CAICT projects that China’s “core” AI industry will surpass 1.2 trillion yuan (~US$170 billion) in 2025, growing nearly 30%, driven in part by rapid adoption of large models in manufacturing, where their share of application cases has risen from 19.9% to 25.9%. ([companies.caixin.com](https://companies.caixin.com/2025-12-15/102393332.html)) The report frames 2025 as a turning point where AI shifts from pure R&D into large‑scale industrial deployment, with 27 specialized data‑collection sites already built nationwide to support embodied‑intelligence training. ([companies.caixin.com](https://companies.caixin.com/2025-12-15/102393332.html)) For global AI players, the numbers underscore how aggressively Chinese cloud providers are scaling inference workloads and point to a home market that could rival the U.S. in sheer usage volume over the next few years.