“Chinese large language models have significantly worse rate limits compared to models from OpenAI and Claude, reflecting underlying compute and capital constraints.”