The AI Chip Owners hub tracks how much computing capacity is held by different entities, broken down by specific chip models, beginning in Q1 2022. These figures include Nvidia and AMD data center GPUs, Google TPUs, Amazon Trainium and Inferentia chips, and Huawei’s AI chips. We estimate that these five categories encompass the vast majority of the world’s dedicated AI computing power. Meta and Microsoft also have their own custom AI chip programs, which are not included here, but they produce far fewer chips than Google’s and Amazon’s. For more details on the methodology used to estimate ownership, see our AI Chip Owners documentation.

We convert chip computing capabilities into H100 equivalents (H100e) based on their relative FLOP/s specifications, specifically their maximum 8-bit specification. However, different chips have different strengths and weaknesses (e.g., memory capacities and bandwidths, software stability). Depending on the use case, an H100e provided by an MI300X chip may be more or less effective than an H100e provided by a TPUv6 or B300 chip. Conversions are thus approximate, and are most accurate for model training, which is typically compute rather than bandwidth-bottlenecked.