Analysis.CastroMedia.org

Description

Amazon operates Amazon Web Services (AWS), a cloud platform that supplies GPU resources for internal and customer AI workloads.

Scope

  • Reports indicate AWS procured around 50,000 NVIDIA H100 GPUs, providing roughly 5×10^19 dense INT8 operations per second.12
  • Older-generation accelerators across AWS regions contribute additional capacity.

Total compute: about 5×10^19 dense INT8 operations per second.

Implications

Amazon’s control over AWS concentrates a large share of global AI compute in one corporation, shaping who can access high-end hardware and at what terms.

Works cited

  1. Observer, “Amazon Orders 50,000 H100 GPUs for AWS,” 2023. https://observer.com/amazon-50000-h100-gpus/ 

  2. Colfax, “NVIDIA H100 Tensor Core GPU,” 2023. https://colfaxresearch.com/nvidia-h100-performance/