ByteDance Is Quietly Taking Over China’s AI Infrastructure

However, artificial intelligence requires not only advanced models. AI demands huge amounts of energy and servers and data centers. ByteDance now uses quite a large portion of such facilities in China.

As reported by SemiAnalysis, in China, there are more than 24 gigawatts of capacity in working data centers. ByteDance occupies about one fifth of that capacity. The firm leases almost all this capacity and does not construct much on its own.

This makes ByteDance one of the most significant clients of Chinese data centers. The increasing collection of ByteDance's AI solutions consists of Doubao, an AI assistant which is frequently used, and Seedance, a video creation platform.

Construction is Booming in China, but AI Chips Are Not Enough

China still has 20 GW of data center capacity under development. Another 30 GW has already been announced. However, building additional data centers does not necessarily mean solving the AI chip problem.

A lot of the existing Chinese data centers were constructed for conventional telecom purposes. These data centers are not equipped for working with the high power density used by current AI chips. Data centers developed for handling AI chip workloads are being filled up much quicker.

The real problem now may be the availability of chips. In the first half of 2026, China increased its AI chip capacity by 595 exaflops. This figure is less than 802 exaflops in the second half of 2025. Analysts blame it on shortages of domestically produced and imported AI chips.

What This Means

This is just one way we believe AI competition will continue to evolve in the coming years. Not only is AI competition a matter of who has better algorithms, but it is also a matter of electricity, chips, and infrastructure.

For you, this means that, in the future, AI services will be even more efficient and powerful than they have ever been before. It also means increasing energy demands and increased computing costs.

If you found this analysis helpful, then please help us by sharing it with your friends. You can also check out some other stories on how the AI infrastructure race continues to evolve.