It should not at all be surprising that running models at home is more expensive than commodity providers. That's just generally true of running your own stuff. Even if the cost in money isn't higher, the cost in time is often _significantly_ higher.

This is why the idea that the AI labs are in trouble because inference will be a commodity is _completely backwards_. Some of the largest and most powerful companies in the world sell commodities. They compete on scale and efficiency, and you are never going to be able to compete with the big labs on either.