Roughly speaking, the best you can do is to harden the server to make it as resistant to compromise as possible.
Ideally, you would store the seed in a hardware security module (HSM; aka a crypto co-processor). You'd ake sure the seed never leaves the HSM, i.e., do all cryptographic computations in the HSM. This offers better protection, though it is admittedly more expensive.
But, as you correctly point out, you cannot store the seed in hashed form. The seed must be stored in the clear, so if that server gets compromised, you are in big trouble. This means it is absolutely vital that you protect that server just as well as you are able to.
OTP seeds are different from passwords. People tend to use the same password on multiple sites; that doesn't happen with OTP seeds. Hashing passwords is used partly to protect users' passwords, so that if site X's database is breached, then X's users' accounts on other sites aren't compromised. That threat simply doesn't apply to OTP seeds.
Also, with passwords, you can hash passwords. If you can, you might as well, since it does help mitigate some risks. (And passwords are so widely used, and used by developers who are not security experts, that it is a near given that many sites who use passwords will experience a security breach at some point.) Since you can't hash OTP seeds, this mitigation simply isn't available for OTP seeds -- so you'll have to use other methods to protect your OTP seeds. Fortunately, only very security-aware sites should be storing their own OTP seeds, so if one is optimistic, one might hope that they are in a better position to apply other defenses.
Anyway, since OTP seeds have different characteristics from passwords, you shouldn't assume that every mitigation for passwords will necessarily transfer over to OTP seeds as well.