CUDA
Nvidia's software platform for programming its GPUs, released in 2007 and a key reason most AI runs on Nvidia.
CUDA lets programmers write code that runs on Nvidia GPUs and comes with libraries for maths, neural networks and communication between chips. Nvidia released it in 2007, years before deep learning took off, and the main AI frameworks were built on it. Its maturity, documentation and the habits of millions of developers form what is often called Nvidia's software moat. Rivals offer their own stacks, such as AMD's ROCm, and tools like PyTorch and Triton aim to make code portable, but moving large workloads off CUDA still takes effort.