When I load an existing HF demo using gr.load(), am I using the original Spaces' hardware to run predictions?

Hi, I have a private HF demo that I load into a public one using gr.load(). It works fine, but I want to upgrade the hardware to get better inference times for my demo. I am confused as to which Space I will upgrade, the private one where the code for the original demo is or the public one which will be the one accessed by users?

I cannot find a clear answer in the documentations: https://www.gradio.app/guides/advanced-interface-features

Thanks!