CUDA Toolkit version 12.6 notes: https://docs.nvidia.com/cuda/nvjpeg/index.html#nvjpeg-decoupled-decode-api

Here I see there are two methods:

NVJPEG_BACKEND_HYBRID - Uses CPU for Huffman decoding.

NVJPEG_BACKEND_GPU_HYBRID - Uses GPU for Huffman decoding. nvjpegDecodeBatched will use GPU decoding for baseline JPEG images with interleaved scan when batch size is greater than 50. The decoupled APIs will use GPU assisted Huffman decoding.

I guess CUDA can do Huffman decoding using: NVJPEG_BACKEND_GPU_HYBRID.

Note: These two methods seems to be not part of the built-in JPEG hardware decoders (which are only found in enterprise GPUs), thus should be done via CUDA cores.