I love how Facebook has solutions to problems that I just would have thrown more money at.
But I wonder if it’s really cheaper to implement and support BOLT than just buy more servers.
Depends on how many servers. Facebook, Google, Amazon et al will gladly throw more hardware at a specific bottleneck. On the other hand, for more systemic kinds of slowness, even a 0.1% improvement translates into far more savings in hardware than the cost in developer salaries. When one cost is tied to scale and the other is relatively constant, a lot of tradeoffs are different for the big players than they are for smaller ones.