I actually think this is a reasonable strategy.
The key thing to get from the fivethirtyeight predictions is that, this close to the election, if the polls are accurate then Trump has no chance. The reason they actually give him a 10% chance is to reflect the possibility that polls are not accurate.
Realistically, in order for Trump to be able to win from this position he needs the polls to be out by at least 3-4%. So it makes sense to assume that is the case - if it isn't, no amount of strategy will make a difference.
But in that scenario, Arizona and Florida (and North Carolina) are no longer marginal Biden states that Trump needs to win. They would be marginal Trump states which still aren't enough for him to win. He would still need at least one more state, which could plausibly be Pennsylvania, Michigan, Wisconsin or Minnesota (Nevada is closer in the polls than some of these but wouldn't be enough on its own). So it might make sense to diversify rather than putting all his eggs in Pennsylvania.
Incidentally, although Minnesota has the longest streak of Democratic voting in presidential elections of any state, that is rather misleading; one of the main reasons it has this record is that it was Walter Mondale's home state (and the only state he carried as Democratic candidate in 1984).