When are Carnot groups complete and negatively curved (in the sense of $CAT(\kappa)$ spaces)?

All Carnot groups are complete metric spaces, since they have all closed balls compact ("proper" metric space). In general, any metric space with a transitive isometry group, and having a compact subset with nonempty interior, is complete (easy exercise).

The result that every Carnot group of dimension $\ge 2$ is not CAT($−\kappa$) for any $\kappa>0$ is straightforward. Indeed, since it has a non-isometric self-homothety, it would imply that it is CAT($−\kappa'$) for every $\kappa'>0$, and hence CAT($-\infty$), which for a geodesic space means an $\mathbf{R}$-tree, which cannot have any subset homeomorphic to the plane.

Actually, a non-abelian Carnot group is not even CAT(0), and does not even have a quasi-isometric embedding into any CAT(0) space (or any uniformly convex Banach space). The latter fact was established in:

Scott D. Pauls. The large scale geometry in nilpotent Lie groups. Commun. Anal. Geom. 9(5), 951-982, 2001. However,

Scott D. Pauls. The large scale geometry in nilpotent Lie groups. Commun. Anal. Geom. 9(5), 951-982, 2001.However, the result that every Carnot group of dimension $\ge 2$ is not CAT($-\kappa$) for $\kappa>0$ is straightforward. Indeed, since it has a non-isometric self-homothety, it would imply that it is CAT($-\kappa'$) for every $\kappa'>0$, and hence CAT($-\infty$), which for a geodesic space means an $\mathbf{R}$-tree, which cannot have any subset homeomorphic to the plane. $\endgroup$ – YCor May 1 at 12:12