It seems to me that a multi-commodity flow problem should be treated differently when the commodity demand constraint is an equivalence and not a lower-bound. Consider a unit demand. Apparently, in such a case, the problem seems easier (and not really a flow problem).
I am not considering the constraint which relates edges with commodities.
My question is whether this intuition is misleading or not. Is the problem still NP-hard?
EDIT: Here's the formulation. For a given graph $G = (V,E)$ with associated capacity function $c : E \rightarrow \mathbb{N}$. Let $K_1, K_2, \dots, K_n$ be a set of commodities, such that $K_i = \{s_i, t_i\}$ -- i.e. a couple of source and sink nodes in $V$ --.
The objective function is a sum over all the flows $f_{e,i} \in \mathbb{N}$.
The first constraint is the flow conservation: $$\sum_{e \in \delta^{-}(v)}f_{e,i} - \sum_{e \in \delta^{+}(v)} f_{e,i} = 0,\ \forall i \in \{1,\dots, n\}\text{ and }\forall v \in V\smallsetminus K_i.$$
The commodity demand constraint is given by: $$\sum_{e \in \delta^{-}(s_i)}f_{e,i} - \sum_{e \in \delta^{+}(s_i)} f_{e,i} = -1,\ \forall i \in \{1,\dots, n\}.$$ $$\sum_{e \in \delta^{-}(t_i)}f_{e,i} - \sum_{e \in \delta^{+}(t_i)} f_{e,i} = +1,\ \forall i \in \{1,\dots, n\}.$$
The final constraint is on the capacity bound
$$\sum_{i = 1}^n{f_{e,i}} \leq c(e),\ \forall e \in E$$
