Story perspectives

China’s AI firms face Nvidia chip bottleneck amid demand

8/21/2026

Summary

Story summary

Chinese AI firms are optimizing software to cope with surging inference demand.
Inference workloads still depend on a limited pool of high-end chips, especially Nvidia processors.
Guan Jiawei, vice-president of Approaching, said high-quality token demand far outstrips supply.
He warned that complex tasks like coding still require Nvidia chips, creating compute constraints.
China’s average daily token calls exceeded 140 trillion in March, over 1,000-fold since early 2024.

1 of 1

Story summary

Chinese AI firms are optimizing software to cope with surging inference demand.
Inference workloads still depend on a limited pool of high-end chips, especially Nvidia processors.
Guan Jiawei, vice-president of Approaching, said high-quality token demand far outstrips supply.
He warned that complex tasks like coding still require Nvidia chips, creating compute constraints.
China’s average daily token calls exceeded 140 trillion in March, over 1,000-fold since early 2024.