We introduce Recursive Inference Scaling (RIS), a plug-in technique that exploits language's fractal structure to boost model performance, unlocking a new dimension of inference scaling.