“At OpenAI, the teams developing text-based models and image/video models (like Sora) operate as separate organizations with distinct roadmaps, productization strategies, and inference stacks.”