I made some comparisons, but I only have an old AMD Ryzen 7 2700X OctaCore CPU with an old GTX 1060 6GB GPU. Generally speaking, using the GPU was always faster than just CPU (of course), and GPU alone was faster than GPU + CPU with CUDA generally faster than OptiX, whereas OptiX with CPU was faster than without. As I said, this could all be dependent on my hardware.
But in all cases Blender was using the GPU when it was enabled. However, there was one discovery that I made: Your performance monitor shows "3D" on the first diagram. I had to switch it to "Cuda" - no matter if Blender was set to CUDA or OptiX - to see the full working load.
This is not a real solution if your GPU actually is not being used, but I simply cannot recreate this issue. As I said, if I switch the diagram to "Cuda", I can see it working in the performance monitor as long as I'm not switching to CPU. Have you switched to CPU intentionally to check against the baking performance with GPU? Is the baking time the same/similar? If not, your Blender will most likely use the GPU.
Another thing which will only use the CPU and always comes between switching from one object to the next is the Denoise for Render. If you have it enabled, after rendering with the GPU, the CPU will take over to denoise the baked image.