It certainly has helped OpenAI and Anthropic get their KV cache costs under control.