Every step of your analysis is several orders of magnitude too optimistic. The only thing sensible here is conceptually going to higher voltage. Loss in these

Every step of your analysis is several orders of magnitude too optimistic. The only thing sensible here is conceptually going to higher voltage. Loss in these systems is dominated by switching, not conduction. Are you just using the Rds(on) figure? It's best to work back from efficiency for back of napkin analysis. At 99 % efficiency (very optimistic for such a large step down ratio), it's 20 kW of loss, not 100 W. That's a lot of power to pull out. (1) Rds(on) Cgs product for a 100 V FET is a lot better than a 10 kV SCR or IGBT. (2) The inductor for a 10 kV, 2 MW converter will be massive, lossy, and hard to insulate. (3) That MOSFET that can only handle a few 10's of W of dissipation before it melts off the PCB, even with forced air/water cooling. Times like these remind me that when people speak so authoritatively on HN they're usually full of shit.