On demand cost is $12.32/GPU/hr. For comparison, FluidStack and Lambda Labs are offering them at $2/GPU/hr on-demand (possibly at a loss or close to it - though

On demand cost is $12.32/GPU/hr. For comparison, FluidStack and Lambda Labs are offering them at $2/GPU/hr on-demand (possibly at a loss or close to it - though AWS isn't doing that - their investors expect significant cloud margins). Even AWS's 3 year reserved is $5.40/GPU/hr which is over 2x the reserved cost at places like Lambda. Cohere, Stability and Hugging Face use AWS [1]. [1]: (I wrote this) https://gpus.llm-utils.org/nvidia-h100-gpus-supply-and-demand/#which-companies-use-which-clouds https://gpus.llm-utils.org/nvidia-h100-gpus-supply-and-deman...