I think the Starcraft pros tend to think about what separates top pros from mediocre pros. They see these things, like better 'mind games', as what makes Starcraft hard.

The thing is, a game needs to be very carefully balanced for these scissor/paper/rock dynamics to emerge. This type of balance is not the default --- it's the product of careful design, and empirical tuning.

In human vs AI, this balance will almost surely be broken. The AI will have perfect control, even with APM constraints. Its attention can be split between two attacks perfectly. It won't play at all like a human.

So, the resulting human vs AI gameplay is unlikely to be strategically deep. If the AI can learn to play Starcraft at all competently, it will probably crush. It will probably find a dominant strategy that it can re-execute exactly the same every time, which the human can't stop.

Making the AI work well for Starcraft will be very difficult. In Chess and Go the actions are discrete; in Starcraft the actions are all analog. But the gameplay requires much deeper search than the Atari games Deepmind learned in its early paper. So I think the game will be very difficult for the AI. But not for the reasons people are talking about.