CUDA
Nvidia's proprietary software platform for programming its GPUs — widely seen as the biggest reason Nvidia's hardware dominance has been so hard for rivals to dislodge.
A GPU on its own is just silicon; CUDA is the layer of tools, libraries, and programming interfaces that lets developers actually put that hardware to work for AI training and inference. Over more than a decade, an entire ecosystem of AI software has been built and optimized specifically against CUDA — which means switching to a competitor's chip often means rewriting or re-tuning a large amount of software, not just swapping a part.
That lock-in is exactly what sits at the center of the antitrust scrutiny Nvidia faces: whether its position rests on CUDA being a genuinely better product that rivals are free to match, or on switching costs high enough to function as a moat regardless of how good competing chips become.