Single core boost can't use much power and thus boost is sustainable forever within power and thermal limits.
Well, that would imply that a core can run at max frequency at max temperature within nominal specifications, i.e. at thermal limit. Cores can achieve higher frequency at lower temperature (about 100 MHz for every 20 degrees C, according to DerBauer).
So, if sufficiently below max nominal temperature, a core can briefly boost to max frequency until the core heats up and forms a hot spot, which it probably does in a fraction of a second. At this point the OS may move the workload to an idle (and cold) core, allowing the hotspot to cool. Or the workload may ease up briefly (e.g. stalling on memory access or I/O), after which the core may have cooled down sufficiently to briefly boost higher again. This is seen as frequency spikes.
It appears AMD compensated for these spikes by lowering the advertised Max Boost frequencies by 50 to 100 MHz for the 1st and 2nd generation Ryzen and introducing the XFR concept accounting for the difference between the Max Boost advertised and the actual max frequency achievable by the CPU at stock settings. The XFR concept seems to have been dropped altogether for the 3rd generation, and the Max Boost now means max (or much closer to it, at least).
To remind everyone about XFR, I'll include this slide again: