Most inference platforms weren't built for custom models. Dedicated Container Inference is. Job-level orchestration, policy-driven traffic control, and hands-on kernel optimization. Customers saw 1.4x–2.6x faster inference on video generation in production.