I bought a PopOS Gazelle with an Nvidia GPU so I could play around with ML stuff. But, looking back on it, it might have been more efficient to just get a GPU i

I bought a PopOS Gazelle with an Nvidia GPU so I could play around with ML stuff. But, looking back on it, it might have been more efficient to just get a GPU instance on AWS or Google Cloud, and just using a remote desktop like this. Anybody tried that configuration? If so, how has your experience been?