If we take HunyuanVideo, which is similar to Sora, as an example, they state that generating a 5-second video requires 5 minutes on 8xH100 GPUs. Therefore, if 10,000 users simultaneously want to generate a 5-second video within the same 5-minute window, you would need 80,000 H100 GPUs, which would cost around 2 billion USD in GPUs alone.