Well, I will say this: we have the technology to do it, today.
We won't do it, probably in the next half-century, and maybe not ever, because of extenuating factors:
1. Mains distribution must be robust. It is exposed to direct lightning strike, nearby (induced) strike, arcing due to crossed wires, tree falls, switching (especially inductive) loads, etc.
2. It must be able to source massive currents, to start rotating machinery, clear fuses, etc. High surge capacity is, in one respect, a passive safety feature.
3. There is another passive safety afforded by AC: arcs are somewhat self-extinguishing. DC arcs are notoriously hard to put out. This reason, alone, may well be sufficient to never adopt a DC distribution system!
AC ground currents also don't cause corrosion -- indeed, it's not even a rare occurrence, as I understand it, for an underground substation (vault) to flood with water -- at least as long as it's not sea water I suppose (but maybe even then!), these can just keep on operating, at higher power loss perhaps, but corrosion is negligible, and not much special material or coating is required of the conductors and switches in these installations.
4. I don't have any idea if the semiconductor cost will come down in the future, or how it can come down in some best-case scenario, and if so by how much -- but at least given current top materials (SiC and GaN), replacing a simple wall switch takes about as much semiconductor solid, plus packaging, to handle the same ratings as a mechanical switch. Similar proportions apply to everything else handling power: relays, breakers, transformers (--> converters), tap switchers (would be obviated by converters, at least), etc.
I suppose I could imagine some solution-based or even biologically mediated route to synthesize SiC crystals in high yield and low defect rate; GaN maaaaybe, but it would take some serious work as neither element is very amenable to anything other than CVD, I think. The goal along these lines would be, maybe they're even terribly slow to grow, but the defect rate and purity are acceptable or even improved, and the energy input is tiny so that they can just be left in a warehouse to grow for months or years even, without the expensive high temperature, high pressure synthesis methods that are required today, so they can end up not only cheaper, but economical to use almost in bulk.
Notably, SiC PVD/sublimation is prone to especially high defect rate (screw dislocations); it's not at all a simple process, and the high temperatures are very energy-intensive to maintain. So if there is a cost-reduction possibility, I can imagine it would be something in the synthesis and processing. But it is something they have managed quite well nowadays, and along with annealing and epitaxy steps, excellent quality substrates are available today; if just pricey.
The chips used in a MOSFET of given rating, are much smaller than Si, so despite the high cost, and give or take the performance advantage, they are economical as replacements for Si MOSFETs. But like I said, you need, just, so many of them to replace something mechanical like a switch, or transformer.
You could synthesize many of the passive safety and practical aspects that AC provides -- but you will do so at great expense and complexity, and the fact that it's a basic mechanically-switched system carrying power around, is a huge asset to the system as a whole, even if you have some annoyances like transformers and PFC correction capacitors (and line reactors, and dealing with harmonics, and..).
And yeah, transformers are generally very efficient; very large (100s MVA) transformers are typically in the 99.something % range. They're huge, the volume/surface area ratio is pulling hard here, so they do need cooling oil to circulate, not to mention to insulate the high voltages, but yeah, they're really great components as things go.
One could also answer your question in the affirmative: we DO have DC grids, silly! -- we just use them where they are most suitable: HVDC links, where the exceptionally low transmission loss at nearly 1MV (a few over 1MV even, IIRC) is worth the cost (double meaning, both capital cost and electrical loss) of the converters at the end points. Being a transmission rather than distribution system, no switching or branching is required, and it can be elevated high above ground to keep the electric field modest.
I suppose corrosion might still be a problem (rain soaked insulators?), but they could add inert electrodes to the support insulators at not too much cost. That is, inert as in platinum or what have you -- electrolytically inert materials are not generally cheap -- but as the overall system is expensive, and relatively few insulators are needed for a wide-span transmission line, it could easily be afforded.
Tim