China now says it has seen a dramatic increase in its domestic compute power for artificial intelligence. The government’s numbers indicate a scale that is out of kilter with the world’s rankings. That has led to speculation of a “dark pool” of compute. The ministry of industry and information technology claims 1,882 exaflops. That’s 1,882 quintillion calculations per second. Compared to the Top500 list, it seems unfair. This is a benchmark for supercomputers. It has China further down the list. The gap is too big. 

But, according to the South China Morning Post, the two measures do not correlate. The Top500 uses more rigorous tests for general computers. China’s figure includes AI workloads. If we correct the figure to Top500 standards, it drops. We get a figure of 120-230 exaflops. But the figure remains high. It’s still higher than publicly reported. It’s closer, but not there.

Beyond Benchmarks: How Different Measurements Shape the Narrative

The development points to an AI system in China that may be more advanced than many had assumed. The full extent remains unclear. In recent years, the country has also stopped sharing details of its strongest machines with global rankings. An expert cited by the South China Morning Post links this shift to rising geopolitical strain and tighter export controls from the United States.

Zhang Yunming said the country is building a nationwide computing network. The design is layered and wide in scope. It aims to support the growth of the AI sector. The system will spread computing power across national and local centres. It should make access easier and costs lower. Smaller firms stand to gain the most.

Global AI Race: How China’s Momentum Stacks Against the US

In the United States, no single figure exists for national AI computing power. Most of the infrastructure sits in private hands, and firms guard their numbers. Estimates from the Stanford Institute for Human-Centred Artificial Intelligence suggest the country still holds between 50 and 75 percent of global capacity. It also leads in the number of advanced data centres. Even so, the distance between the two sides seems to shrink. A recent Stanford report notes that Chinese AI models now rival those from the US. The progress is quick and plain to see.

Projections add weight to this view. A study by International Data Corporation and Inspur points to steady growth in China’s AI compute. It places the annual rate at about 46 percent from 2023 to 2028. That pace is more than twice the rate seen in general computing.

Final Words

Ultimately, the AI compute narrative in China is not so much about who has the most but is a lesson that metrics, like mirrors, are as prone to distortion as they are to reflection. Take away the exaflops in the headlines and you are left with an impressive piece of work: a fast-growing ecosystem, a nationwide network being built and an unmistakable desire to democratize access to computing power. Meanwhile, the United States is not exactly resting on its laurels, as its hegemony is still based on the power of the private sector and hi-tech infrastructure.