Proof of Work can be done in a way that is extremely expensive to scale - see Argon2 or RandomX algorithms. Basically, the idea is to resist ability to offload computation to GPU/FPGA/ASIC. Without them and with a memory-intensive algorithm, cost of an attack grows fast.
That is the folklore, but theres no reason you couldn’t make an ASIC for any “asic resistant “ algo
Yes, but in the meantime whilst they plan out their fabrication for such an ASIC we can have a win. So, I don't think it is a bad take. We're making this spammer sound like an overlord with abundant resources.
ASIC are good as long as you don't need hefty amount of RAM, as it is for Argon2. See this paper: https://www.cryptolux.org/images/0/0d/Argon2.pdf Basically, the point of memory-hard hash function is to make parallelization done by ASIC expensive due to memory usage.