By the end of 2027, China is projected to increase its AI computing capacity from 4.6 million to 8.8 million H100 equivalents. According to the Institute for Progress, approximately 65% of this capacity will rely on American-designed chips from companies such as Nvidia and AMD.
The report highlights that the United States currently maintains a 16-fold advantage in computing resources for AI training. Despite this, Chinese AI models are estimated to lag behind their American counterparts by only 6.5 months, partly due to the use of model distillation techniques.
To address these trends, the Institute for Progress has urged US authorities to tighten controls on chip smuggling and the legal export of AI hardware. Analysts suggest that foreign entities currently bypass direct purchase bans by leasing cloud computing power from US providers.
Huawei is working to expand domestic alternatives, with its Ascend 960DT chip expected in the first quarter of 2027. The company has also deployed over 1,000 Atlas 900 A3 SuperPod clusters, though it faces significant supply constraints related to HBM memory availability.
