If you're asking this question because you want something easier to implement than a Delaunay triangulation algorithm you're most likely out of luck.
You should also specify in what space you're operating. If you're working with $\mathbb N^2$ coordinates, you can possibly do it in $\mathcal O(n \log\log n)$ (with a randomized algorithm to generate the triangulation).
If you have $P\in \mathbb R^2$ your choices (ordered by difficulty of implementation from easiest to not-easiest) are:
- Delaunay triangulation
- WSPD based approaches (first find clusters on Quadtree of $P$, then build MSTs for each cluster and then connect clusters to get complete MST)
- Improved version of Boruvka's algorithm: paper
The last one is also an improvement over the other two in higher dimensions. This answer is more or less a condensed version of the literature review of the linked paper.