While the first mover advantage exists, I don't think it's a sufficient moat in the long term.

OpenAI's moat is the high cost of training as you mentioned but this might be obsolete in a few papers down the road.

They themselves realized this by trying to turn it into a platform but I don't think it's enough.