To echo the statements made here already with a real world example I just had a few months ago, I had (2) 450HP / 480V motors being driven from VFDs with maybe 50-75VA of other auxiliary loads at a site. 450HP is 515A from table. We were concerned that a 1MVA transformer might be cutting it too close to supply all of this and also left no room for growth, so we up-sized it all the way to 1.5MVA. It turns out we would have probably been just fine at 1MVA based on actual startup and running current measurements during commissioning. I don't think we ever even break 1000A (~830kVA) at full load, and starting is nothing but a smooth ramp up in current.
Our motors were driving centrifugal pumps with a very friendly ramp up time in the VFD, which goes to Jraef's point about the VFD, load, and starting characteristic playing a role in virtually eliminating the starting current when using a VFD in certain applications.