China's use of artificial intelligence is scaling at a startling pace. According to Minxiao Chang of the South China Morning Post, the country's National Data Administration says daily AI "token" consumption reached 140 trillion in March 2026, up from 100 trillion in December 2025. Both figures dwarf the roughly 100 billion tokens consumed daily in early 2024.
Tokens are the small chunks of text an AI model processes to read a prompt and write a response, so token volume is a rough proxy for how much AI work a country is actually doing. To put the growth in human terms, the report cites a Beijing-based ByteDance employee who says he personally burns through close to a billion units every month.
The surge coincides with Chinese labs pushing more capable models into the open. Crypto Briefing reports that Moonshot's newly released Kimi K3 has stunned observers with 2.8 trillion parameters and what it calls competitive pricing — a combination of scale and low cost that helps explain why usage is climbing so fast.
The quality gap is narrowing too. According to the AI Security Institute, recent open-weight models now trail the best closed, frontier models on cyber capabilities by roughly four to seven months. That is a tighter margin than the six-to-ten-month gap the institute tracked through most of 2025.
Taken together, the three data points sketch a single trend: cheaper, more openly available Chinese models are getting closer to the cutting edge, and people are using them far more.
Why it matters: when powerful AI becomes abundant and inexpensive, adoption can outrun the safeguards and oversight built for a slower, more exclusive era.