ByteDance increases its purchases of domestically produced chips, and major internet companies are racing to build computing power moats.

The progress of ByteDance's data center construction has new updates, according to industry sources, ByteDance is in discussions with Tianshu ZhiXin to purchase at least 50,000 AI chips, mainly for inference work. It is understood that the chips to be supplied in this negotiation are mainly used for large model inference workloads, corresponding to Tianshu ZhiXin's ZhiKai series cloud inference GPU, while training scenarios use the Tiange series. If the deal is reached, Tianshu ZhiXin will become ByteDance's third GPU supplier after Huawei and Cambricon. As of the time of writing, there has been no response from either ByteDance or Tianshu ZhiXin. Internet giants represented by ByteDance are ramping up their own data centers, not just to address short-term emergency computing power gaps, but to reflect deep industry changes: the structural shift in AI computing power demands, the continuous upgrade of enterprises' strategy towards self-controllable computing power, and the concurrent ground deployment of training and inference hardware along with the localization of domestic computing power are entering a landing phase.