Which company has the strongest AI computing power? Take a look at the data center ranking list.

Research institution @EpochAIResearchI released a global AI data center TOP60 ranking as of July 24.
To ensure fair comparison, regardless of what chips are used, everything was converted into equivalent computing power of NVIDIA H100.
The results show that currently, the top three AI data centers by computing power are all in the United States:
Elon Musk’s Colossus 2 in Memphis, Texas ranks first, with the equivalent of 1.11 million H100s, built at a cost of $35.8 billion, far ahead of the rest.
Microsoft in Atlanta and Meta in Ohio follow closely, both at around 770k H100 equivalents. Notably, OpenAI’s Stargate in Texas currently stands at fifth place with the equivalent of 510k H100s, but it plans to double by year-end to 1.02 million, when it will be right on the heels of the No. 1 spot.
Among the TOP60, Google has 16 entries and Meta has 11, basically covering the power “valleys” in the U.S. Midwest and Southeast.
China has three entries, all in the north:
Huawe*’s Hologring data center in Hohhot, Inner Mongolia ranks 37th;
CenturyLink’s (世纪互联) Ulanqab data center in Ulanqab, Inner Mongolia ranks 39th (it is speculated that it mainly serves ByteDance),
and Alibaba’s data center in Zhangbei, Hebei ranks 54th.
How big is the gap?
China’s highest-ranked Huawei has equivalent computing power that is one-ninth that of the No. 1 Musk Colossus 2. Alibaba’s Zhangbei site actually installs 210k H20s, but each H20’s computing power is only 15% of an H100’s, meaning the per-card computing power gap is an order of magnitude.
Hua*’s Ascend 950 and Alibaba’s Zhenwu 890 are both on the way; in the future, this ranking will definitely not remain this pattern for long.
The AI race is not only about models, but also about infrastructure; and the infrastructure gap is harder to catch up than the model gap.