Kernel launch refers to the process of starting a computation on a GPU by invoking a kernel function from the host CPU, specifying the number of threads and blocks to be executed. This operation is crucial for parallel processing in GPU programming, enabling efficient execution of data-parallel tasks across multiple cores.