A growing number of Chinese artificial intelligence labs are experimenting with training their models on homegrown chips instead of relying entirely on Nvidia's hardware, according to the South China Morning Post.
The shift is still partial. Per the South China Morning Post, the labs are moving earlier phases of model training onto domestic silicon — a notable step, since training is the most chip-intensive and demanding stage of building an AI system. The report points to five Chinese AI models that have been trained using local chips as evidence that the approach is gaining traction.
The broader backdrop, as the South China Morning Post notes, is that Chinese AI models have become increasingly competitive. That raises a pointed question the outlet puts directly: can Chinese silicon actually replace Nvidia, the company whose processors have become the default engine for cutting-edge AI development worldwide?
The sources here describe experimentation and momentum rather than a finished replacement. They do not claim Chinese chips have matched Nvidia's performance, only that labs are testing how far domestic alternatives can carry the workload.
Why it matters: Nvidia's chips sit at the heart of the global AI race, and U.S. export controls have limited China's access to the most advanced ones — so any sign that Chinese labs can train competitive models on their own silicon signals a potential loosening of that dependence and a reshaping of who controls the hardware behind frontier AI.