Caruana comes heavily armed with prep against Karjakin's pet QID and puts Black into unknown territory, gaining a lot of time meanwhile. Instead of continuing "fluid" play where many options are possible (and Caruana may know some tricks along the way), Karjakin decides that he needs to go for forcing lines into a position he believes he can play. As it happens, this possibility involves queen sacrifice and very deep calculation; it is clear that Black puts his hopes on a fortress-like position, where he was even prepared to give away some pawns in exchange for stabilizing.

White manages to maintain uncomfortable pressure through the complications and wrecks the Black pawn structure on the kingside. He continues with 28.Qc3, threatening the deadly 29.g5, and at this point Black takes another risky decision. I think he was always prepared to part with the d-pawn as well, and now there was the chance to force further exchanges.

30.g5 has been much criticized, mainly because the SF engine on Chessbomb gives it a red colour, indicating a blunder. But it is clear that White considered the natural alternatives (30.Bf3 is the move most people would play in blitz) and that he decided the resulting position was a fortress. If Caruana had played into such a line, made 15 or so more moves and conceded the draw, most likely he would have been criticized for <not> playing 30.g5, as this at least offered practical chances. I think 30.g5 is a well-considered winning attempt and as such it demands a lot of respect.