Domestic Chip Data Centers Go Live: AI Compute Autonomy Enters 'Hardcore' Phase

In 2024, China's AI chip imports dropped by about 70% year-on-year, yet the average compute utilization rate of domestic replacement chips was only around 60%. Against this backdrop, Zhipu AI announced the construction of a data center fully based on domestic chips, using Huawei Ascend and other local solutions for everything from training to inference. As a first-year master's student who just joined the lab, I feel both excited and confused by this news—excited because domestic computing power finally has a benchmark for large-scale implementation, but confused because can this "purely domestic" route actually make large models work within the current ecosystem?