CUDA is the software layer that connects AI code to Nvidia's hardware. Because most AI tools and libraries were built and optimized for CUDA over more than a decade, switching to a rival chip has meant rewriting and re-tuning software, a cost that has kept customers on Nvidia even when competitors' chips were cheaper. Anything that makes it easier to write AI software for non-Nvidia chips, such as AI coding assistants, chips away at this lock-in.