China’s AI token economy surges as MiniMax and Zhipu post big gains

This piece makes explicit what’s been building quietly for two years: China is not just training big models, it’s turning itself into a ‘token factory’ with massive, monetized usage. MiniMax and Zhipu aren’t talking about abstract benchmarks—they’re reporting triple‑digit revenue growth and huge token sales, much of it from overseas customers. Pair that with Moonshot’s Kimi K2.5 powering a U.S. coding product like Cursor, and you get a picture of Chinese models woven directly into Western software stacks despite the geopolitical noise.

The other half of the story is infrastructure. Beijing is tying this token boom to concrete plans for hyperscale intelligent computing clusters and power‑aligned data centers, effectively nationalizing part of the compute layer needed for frontier training and large‑scale inference. If China really can couple abundant relatively cheap clean energy with vertically integrated AI hardware and models, it can keep price‑performance advantages even if individual models occasionally lag the West on peak benchmarks.

For the race to AGI, the implication is that ‘capability’ is increasingly a function of deployed tokens, not just model cards. A country that can run orders of magnitude more agentic workloads—because its models are cheaper and its energy base is prepared—will iterate faster on products, safety techniques and emergent behaviors. The Xinhua framing of China as a future ‘token factory’ is less rhetoric than a blueprint for compounding advantage.