Ben Thompson’s daily update today about “thin vs. thick” AI continues to sound right to me:

… you can envision a world where compute is no longer scarce, and very cheap models in the cloud are also good enough; it’s hard even then to see local inference penciling out economically. That’s why the true case for this setup may ultimately depend not on free as in beer but free as in speech…

Cloud AI is going to be best for most people. Local AI is great not because of cost but for control.