Dolphin society is typically described in the same way as primate society. Fission-fusion societies among dolphins are characterized by two levels of social hierarchy: groups of two or three related males (“first-order alliances”) work together to guard one or more females from other males. Those first-order alliances combine to form larger groups (“second-order alliances”) which cooperate to “steal” females from other groups first- and second-order alliances. Since the individuals in first- and second-order alliances are related and therefore share genes, this sort of cooperation can be explained by kin selection.
Research that was published in 2010 described a new kind of dolphin group formation: a set of alliances among second-order alliances, or "second-order super-alliances." These third-order alliances are comprised of individuals who are not related, so these super-alliances can not be explained by kin selection. It also can't be described by reciprocal altruism, because previous interactions did not sufficiently predict future interactions (in other words, they are not organized on a tit-for-tat system). It turns out that second-order super-alliances are incredibly familiar to human friend groups: each dolphin ranks the relative importance of his friendships with other dolphins, but that is balanced by his prediction of how his dolphin friends rank him as well! It is that sort of mental calculus that helps a dolphin decide who to help and when to help them.
Second-order super-alliances are made of, on average, around twenty-five males. But a pod of over 2000 individuals? I'm not sure what explains that, but it seems to be rare.