Analysis.CastroMedia.org

Description

Microsoft operates Azure, a global cloud platform providing large-scale GPU clusters for internal and customer AI workloads.

Scope

  • Reports indicate Microsoft acquired around 150,000 NVIDIA H100 GPUs for Azure AI and OpenAI workloads, offering on the order of 1×10^20 dense INT8 operations per second.12
  • Additional GPUs across Azure regions and classified government clouds further expand capacity.

Overall, Microsoft’s accessible AI compute is roughly 1×10^20 dense INT8 operations per second.

Implications

This scale allows Microsoft to serve commercial customers and partners like OpenAI, but concentrates critical AI infrastructure under one corporation.

Works cited

  1. Observer, “Microsoft’s Massive GPU Purchases Fuel AI Race,” 2023. https://observer.com/microsoft-150000-h100-gpus/ 

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