China's AI Compute Power Soars, Revealing Vast Hidden Capacity
China has announced a staggering leap in its domestic artificial intelligence (AI) computing power, with official figures indicating capacity that vastly surpasses what is reflected in global benchmarks. This revelation has sparked questions about a potential "dark pool" of hidden compute resources, suggesting China's AI infrastructure may be more advanced than previously understood.
Massive Discrepancy in Reported Figures
According to China's Ministry of Industry and Information Technology (MIIT), the country has achieved 1,882 exaflops of computing power, equivalent to 1,882 quintillion calculations per second. This figure is more than 6,000 times higher than China's computing power recorded in the Top500 list, a widely used global ranking of supercomputers. Experts note that the two metrics are not directly comparable, as the Top500 uses stricter standards for general-purpose computing, while China's figure is based on AI-focused calculations, which count simpler operations and yield significantly higher performance numbers.
When adjusted to Top500 standards, China's capacity would fall to approximately 120 to 230 exaflops, still far above publicly known levels. This discrepancy highlights the different methodologies used in measuring computing power and underscores the complexity of assessing global AI capabilities.
Geopolitical Tensions and Infrastructure Development
The development comes amid rising geopolitical tensions and US export controls, with China having stopped submitting details of its most powerful systems to global rankings. MIIT Vice-Minister Zhang Yunming stated that China is building a nationwide, multilayered computing grid to support its AI industry. This system aims to distribute computing power across national and local centres, making it accessible and affordable, especially for smaller businesses, thereby fostering innovation and growth in the sector.
US Comparison and Narrowing Gap
In contrast, the United States does not publish a single national figure for AI computing power, as most infrastructure is privately owned. Estimates from the Stanford Institute for Human-Centred Artificial Intelligence suggest the US still holds 50–75% of global AI capacity and leads in the number of advanced data centres. However, the gap between the two countries appears to be narrowing rapidly.
A recent Stanford report found that Chinese AI models are now competing closely with their US counterparts, signalling rapid progress in capability. Growth projections further underline this trend, with a report by International Data Corporation and Inspur estimating China's AI computing capacity will grow at an annual rate of 46% between 2023 and 2028—more than double the pace of general computing expansion.
This rapid growth indicates China's aggressive push to dominate the AI landscape, potentially reshaping global technological dynamics. The hidden capacity revealed by the MIIT figures suggests that China's AI ambitions are backed by substantial, albeit less visible, infrastructure, raising strategic concerns for other nations.
