The nice thing about local AI is that it really can run anywhere, you just need enough storage space for the weights and the context. It just gets slower if you run it on potato-level hardware.

For bang-on user cases like coding, sure. For concept art and other still-image diffusion tasks, sure. For damn near anything else, calling hardware that doesn’t approach the inference breadth and performance of data-center-hosted remote services ‘potato-level’ is pretty disingenuous… never mind any significant training. Not only that, hardware is less accessible than it’s been in years— nvidia is re-releasing the 3060 so gamers can buy something. For anything you’d stake your business on? Good luck.