Ordinary relays are rated for a few kV transient, in that they have to deal with mains voltages, surge included. But you won't ever find such a DC rating, and only see it attested rarely as such, with a "250VAC" or whatever rating being often the most you will ever see (digging into datasheets or appnotes, you may uncover category rating or the like, which determines surge rating).
So, one could make such an assumption, and proceed. Hopefully just for a one-off sort of thing, with adequate shielding and protection in case switching occurs under load, and arcing and ignition occurs in all relevant relays.
But that's not a great assumption to make.
Better is to get proper relays. They will be quite specialty. Notice you aren't shopping for DC relays, necessarily: those involve special ratings to open and close at up to the ratings. The voltage and current are not so much intrinsic limits of the contacts themselves, but a type rating based on underlying assumptions involving a range of load characteristics (inductive, resistive, inrush, etc.) and the switching thereof. All the design, testing and certification that goes into a given relay, is justified by its mass-market production: make a few hundred thousand, or billion, of something and you can pay off that NRE pretty well. Only making some thousands of some specialty part, though, the sale prices go waaaay up, even if the certifications and such go down.
So, you could ask mfg reps if they can recommend a part, or a part under alternate use given the specific operating conditions of your system, or perhaps even have them come up with something, not necessarily a customized part as such, but just that it's had additional back-end engineering done to check it out for such a purpose only -- but you will likely find few results with the former, and great expense in the latter, if any at all.
It's probably not quite as hard these days, as medium voltage DC applications become mass-market: solar arrays have comparable ratings, for example. (Is this perhaps even a solar panel simulator application?) And, keep in mind there are minor manufacturers, and traditional sales channels, offering somewhat-specialty industrial products in low to modest quantities (thousands/yr ballpark). Usually regional, so, I can't exactly make a suggestion, but they are out there. You might be able to fish around some of those suppliers -- their salespeople are quite knowledgeable about their suppliers, and have close contacts with their reps/FAEs.
As for the overall project, whether such a solution is more economical overall -- in more general terms that is, such as project timeline/completion, or operating costs -- compared to the identified solutions (MOSFET/IGBT, give or take heatsinking), that's something you'll have to figure out.
I will strongly encourage mechanical contacts: do not ignore fault conditions. Inevitably something's going to short out the output, and all your supplies' capacitance discharges through the poor switches. In a series stack, some will be reversed (depending on their relative ESR and capacitance, or inductance even), and then you have to handle ungodly currents in reverse as well; fusing or (supply) control doesn't do you anything because we're talking ~ms time scales here. MOSFETs are useless for such (100s, 1000s A) currents; they can be turned off fast enough to protect themselves (10s us), but keep in mind the necessary overhead to handle the corresponding transient voltages plus clamping into a TVS (MOSFETs are rarely avalanche rated to high currents, and rarer still, repetitive avalanche), which also needs to be high enough to handle whatever expected or external surge/overvoltage might occur (whether because of switching or external surge). Mechanical contacts are extremely robust: the on/off ratio is literally unbeatable, so it should be no surprise that the surge capacity (current when on, voltage when off) is similarly excellent.
Also not to mention, keep in mind the possibility that such transients (~kV sparking) knock out the CANbus or what have you. At least momentarily (wasting a retransmit cycle), if not persistently (outright damaged transceivers). (The latter should at least be easily avoided with appropriate protection and isolation, but that does need to be verified to some extent e.g. ESD and surge testing.)
IMHO this setup is better build using an adjustable power supply (or several in series) suitable to work up to 2kV. One of the major problems I see is sourcing relays that can deal with breaking 2000V DC at several amps.
Breaking isn't required.
Tim