Cost, and the fact that it doesn't actually buy you much. I've evaluated GaN for a couple recent designs I've been involved in and found that its benefits simply didn't justify its costs.

What GaN actually does is, mainly, decrease switching losses. That means switching frequency can go up, or efficiency, or power density. These are all good things, to be sure, but the magnitude of improvement GaN brings to most designs is simply not large enough to justify the cost of the parts and the increased design attention needed to use them. (They're fussy little things.)

That said, I'm bullish on GaN in the medium-term. There are some cool tricks that are infeasible or impossible without them, and they really are better in a lot of ways. It'll just take a while for them to trickle downmarket, and even still they're not going to be replacing Si FETs anytime soon. (Part of the lag in GaN adoption is simply that Si FETs have gotten really, really good.)