Each iteration processes all edges. You run iterations. Total: . In dense graphs where , this becomes . Dijkstra with a binary heap runs in , which is faster when edges are non-negative. For sparse graphs where , Dijkstra is while Bellman-Ford is .
Bellman-Ford trades speed for correctness with negative weights. Use it only when necessary. If all weights are non-negative, Dijkstra is always the better choice.
Space complexity is for the data structures used.