OpenAI's Orion (GPT 5/Next) is partially trained on synthetic data generated with a large version of o1. Which means if that works the data scarcity issue is more or less solved.