The projected growth and the weight of US chips
The artificial intelligence computing capacity accessible to Chinese entities is set to double in 2027, rising from about 4.6 million to 8.8 million "H100 equivalents", the unit of measure based on Nvidia's data-center GPUs. That is according to research by the Institute for Progress (IFP), a Washington-based think tank, reported by the South China Morning Post.
The report's central finding concerns the origin of this capacity: about 65% of total capacity would be ensured by US-designed chips. An indication, according to the research, of China's persistent dependence on foreign hardware even during a phase of rapid expansion of its computing resources.
The share of remote access, that is, capacity obtained through clouds located outside national borders, would grow even faster: from about 1 million H100 equivalents this year to 2.8 million by the end of 2027, according to the estimates cited.
The gap with the United States and the distance between models
The report places China's growth in a comparative framework: the United States would retain a lead of about 16 times in computing power dedicated to training artificial intelligence models. This is a quantitative primacy that, according to the IFP, does not translate into an equivalent gap in results.
China's leading models, again according to the research, would be only 6.5 months behind their American competitors. The combination of the two figures — a very wide compute gap and a limited time lag between models — is the most significant element of the report, because it suggests that the availability of computing power is not the only factor determining the pace of development.
It should be noted that these are estimates contained in a research document: the available material does not specify the methodology used to calculate the H100-equivalents metric, nor how chip design is attributed.
What remains to be verified
The report does not clarify, at least in the available portion, which remote-access channels drive the projected growth, nor how the share of US-designed chips is reconciled with current export restrictions, a topic the material does not address. The margins of error of the 2027 projections are not indicated either.
Nor are there any statements from Chinese or US companies or authorities in response to the research. The confirmed elements are therefore the figures and conclusions attributed to the IFP; their interpretation — particularly regarding the effectiveness of export-control policies — remains a matter of assessments that the material provided does not cover.
