it's going to be quite slower than remote providers, won't be able to run full models, only quantised, and for more cost than the highest Claude subscription tier where you can do roughly 60-100M token per month at 200$, e.g. 7200 for three years. It makes sense if you prioritize local AI but not really otherwise sadly.