My thought is that the 52-43 margin in 2016 was unique to Trump's unpopularity with white college-educated Texas suburbanites who are typically conservative. The question for 2018 is whether they'll vote like normal Republicans and pick Cruz, or whether they'll treat Cruz like Trump and vote for O'Rourke.
Just saw that Enten analysis that says Democrats' major gains are in Obama-Trump areas, or at least areas of that trend, not Romney-Clinton. So I wouldn't get hopes up for O'Rourke, but Democrats should feel better about some of their vulnerable Trump state incumbents.
I don't think the election results we've seen bare that out, tbh. Don't get me wrong, I'm sure there have been huge swings in some Obama-Trump areas too and I'm pretty bullish about the chances of a huge Democratic wave in the midwest (including the rust belt*), but the biggest swings seem to be in affluent, well-educated suburbs regardless of how they voted in 2012 and 2016.
NoVA is the most obvious example (although the Republicans also took a beating in the Virginia Beach area, among other places) and one could attribute the massive drop in suburban Republican turnout in the Birmingham suburbs to the freak circumstances of the Alabama special election, but I think the most important swing we saw last November came in the local elections in the Philly suburbs. The Republicans were losing local countywide offices (sheriff, controller, recorder of deeds, etc) left and right in places like Chester County, Bucks County, and DelCo which is almost unheard of; which (along with some other polling and election results such as the number of popular Republican incumbents who went down in NoVA) suggests that many suburban voters aren't really distinguishing between Trump and anyone else with an R next to their name. At the same time, the Republicans are also seeing huge under-performances in special elections in Obama-Trump areas and even (I assume) places like PA-18 (although we technically won't know for sure until next week).
Ultimately, the Republicans have likely lost an unprecedented number of suburban voters since Trump was sworn in (including Romney-Trump, Obama-Trump, and Romney-Clinton voters), but they're also not really making up that lost ground anywhere else. All signs suggest that the Midwest is about to swing back to the Democrats in a big way (with the Obama-Trump voters there being pretty fed up with the Republican Party in general) and that while some of the WWC folks have officially become Republicans, not nearly enough new ones** are doing so to stop the bleeding, much less actually create a coalition that will be viable in the long-run. Whether it comes in 2018, 2020, 2022, or 2024, the Republicans are drawing closer and closer to a long stretch of time in the political wilderness (assuming the Democrats don't completely f*** it up, but no way to predict whether that'll happen).
You can only game the rules for so long before your luck runs out; at a certain point political parties in a two-party system need a national coalition which can actually win the most votes in order to remain viable in the long-term. The Republicans didn't have one in 2012, they didn't have one in 2016, and they definitely don't have one now that the party is about as popular as leprosy with even the suburban voters who held their noses and voted for Trump in 2016.
*Speaking of the rust belt, as long as the Democrats don't nominate Warren, Booker, or Harris (who would be a Hillary-level weak candidate against Trump), I think we'll see a big swing against Trump in the rust belt in 2020. Maybe a realignment did happen there, but one election between two historically horrible candidates in a "mad as Hell" cycle does not a realignment make. Unlike the suburban shift, I don't think we've seen all that much evidence since Trump took office that what happened in the rust belt was more than a one-cycle fluke. Don't get me wrong, 2016 was arguably the culmination of a smaller realignment in coal country, but that had been underway for a few cycles already. I don't think we've seen much evidence that Trump added very many voters to the Republican coalition in the long-term. A lot of the stuff in Appalachia likely would've happened anyway and Biden would've crushed Trump in the rust belt, I think.
**A lot of these folks haven't voted Democratic in quite some time, so it seems kinda misleading to call them newly registered Republicans
