In general, specialized hardware tends to be expensive (niche uses result in poor economies of scale), which makes it difficult to see more widespread adoption,

In general, specialized hardware tends to be expensive (niche uses result in poor economies of scale), which makes it difficult to see more widespread adoption, and consequently makes them less usable by researchers. If these types of hardware end up finding use in a new problem space (like bitcoin mining), then in theory that could make them more popular, and thereby cheaper (as well as increasing the acceleration of the technology, leading to better, faster, hardware for everyone who uses them). So in theory, even if Bitcoin mining doesn't have any "direct good" that comes from it, it could still be indirectly helpful to the larger scientific computing community.