Former OpenAI Researcher Targets $100 Billion RL Data Market

Competition among large language models is shifting beyond parameters and computing power toward high-quality data that can improve reasoning. Former OpenAI researcher Andrew Ho says the shortage of rigorous reinforcement-learning, or RL, datasets has become a key constraint for frontier systems, potentially turning specialized training data into a major new segment of the artificial-intelligence supply chain.

Ho recently said he had left OpenAI to start a company focused on producing high-quality RL training datasets for complex reasoning tasks. He predicts frontier AI laboratories will spend more than $100 billion on data over the coming years. Publicly available details have not specified the venture’s name, funding, or exact founding date.