ByteDance increases its purchases of domestically produced chips; major internet companies race to build a computing power moat.

According to ME AI , ByteDance's data center construction has seen further progress. Industry insiders say ByteDance is in talks with Tianshu Intelligent Chip to purchase at least 50,000 AI chips, primarily for inference tasks. It is understood that the chips being negotiated will mainly be used for large-scale model inference workloads, corresponding to Tianshu Intelligent Chip's Zhikai series cloud inference GPUs, while the Tiangai series will be used for training scenarios. If the deal goes through, Tianshu Intelligent Chip will become ByteDance's third GPU supplier, after Huawei and Cambricon. As of press time, neither ByteDance nor Tianshu Intelligent Chip has responded. The increased investment in self-built data centers by major internet companies like ByteDance is not merely a short-term emergency measure to fill computing power gaps, but rather reflects a profound industry transformation: a structural shift in AI computing power demand, a continuous upgrade of enterprise self-sufficiency strategies, and the simultaneous implementation of separate training and inference hardware deployments and large-scale replacement of domestic computing power. (Source: ME)