Because engineers are lusting for 1000 tokens per second. You can only achieve something like that with OpenAI.