The OpenGL pipeline has been designed to be data parallel.  For example, processing a fragment can be done independently of other fragments.  Therefore, a shader program can be run in parallel for each fragment, running simultaneously on different threads.

Modern GPUs have a massive number of cores. For example, the NVidia GeForce GTX 980 has 2048 CUDA cores.  Each core can do vertex, geometry, or fragment shader calculations.

Although shaders were originally designed for graphics, their parallel nature combined with performance of GPUs have found new applications beyond graphics.  General purpose GPU (GP-GPU) programming is an interesting topic; different programming mechanisms exist (compute shaders, OpenCL, etc.).

High