There are some competitions that are somewhat similar. - Visually, generals.io is pretty similar to Halite-I, but in actuality ends up being a pretty different game. - Halite-I is most similar to the Ants AI Challenge of 2011 (which inspired it). - Halite-II was conceived of as sort of a mix of Halite-I and Planet Wars. - MIT Battlecode is similar in controlling lots of pieces, but the style of game is very different because Halite is all about emergent behaviors in the game whereas traditionally Battlecode hasn't been. They're also very different from the implementation side for the end-user. - Codingame can range from somewhat similar to very, very different.
Any game where the last survivor wins are based on the idea to get involved as little as possible early on (unless extremely high rewards are given along the way).
Regarding your point about the last survivor winning, that's partially true, but partially not. What you're essentially describing is a proto-version of the non-aggression pact we saw develop in the last weeks of Halite 2016 (and which was indeed largely successful). However, even there, it wasn't quite so simple; competition winner mzotkiew had an excellent write-up of his strategy at https://github.com/mzotkiew/HaliteBot/blob/master/writeup.pd.... Halite-II makes non-aggression even more difficult due to the lack of discretization found in Halite I.
source: conceived of and co-developed Halite.
Transgressing the Boundaries: Towards a Transformative Hermeneutics of Quantum Gravity
https://physics.nyu.edu/faculty/sokal/transgress_v2/transgre...