Overview
- CCTV Finance, cited by ITHome, reports daily token calls in China have exceeded 140 trillion after a surge of more than 100,000%.
- New large models and fast adoption of AI agents are driving the spike in usage, pushing more workloads to rented GPU clusters.
- Industry experts say demand outpaces supply, and they expect the compute rental market to keep growing in the near term.
- CCID’s Xu Zifan says the sector is highly prosperous yet structurally tight, with high‑end GPU utilization above 90% and a 2026 market size forecast of 260 billion yuan.
- A token is the smallest unit of text used to run and bill AI models, so rising token counts signal heavier compute use that can mean longer queues and higher costs for developers and businesses.