Right, high voltage parts are slow as molasses. They have to be, it's just physics.
There are a lot of old machines, still around to this day, that used stacks of 200V SCRs (operating at, say, ~650V DC link, they'd need at least four in series for each switch in the inverter), because they're faster than the higher voltage parts. One example I've seen, a Radyne Ltd. power supply (ugly green paint -- classic 70s industrial!) driving an induction heating load at 3rd harmonic for ~40kHz output. Used small (37mm?) "hockey puck" devices to deliver something like 15kW.
So, you may find it's advantageous to go down rather than up.
Likewise for IGBTs, Stanford and others (or maybe not, I forget who all exactly) have some papers about extracting maximum peak power, how far the SOA can be pushed and what help can be drawn from overdriven gate voltage.
Offhand, I've seen quite peppy IGBTs for the plasma driver market. They're the lowest voltage that's even practical in an IGBT, usually 330V (since any lower and a regular MOSFET is better). A TO-3P rated for 500A peak is not bad at all. These may be losing availability as PDPs fade back into obscurity.
Win Hill (of Harvard and AoE fame) is currently experimenting with SiC MOSFETs for high voltage ~(single digit)ns edges. A few dozen in parallel will get you there.
Tim