Usually a CAT(0) group is defined to be a group acting properly isometrically and cocompactly on a CAT(0) space, but I would like to consider only those groups that act properly, isometrically and cocompactly on a finite-dimensional CAT(0) space.
So is there a group I have to leave out?
Not every CAT(0) space with a proper isometric cocompact group action is finite-dimensional. For example the trivial group acts on the compact CAT(0)-space $[0;1]^\mathbb{N}$.
First, I suppose that by proper action you mean the one in the sense of Bridson and Haefliger, otherwise you would have to regard ${\mathbb R}$ as a $CAT(0)$ groups. Now, it follows from Eric Swenson's paper "A cut point theorem for CAT(0) groups" (Journal of Diff. Geometry, 1999) that the ideal boundary of the $CAT(0)$ space $X$ (on which a group $G$ acts geometrically) is finite-dimensional. This suffices for many practical purposes. For instance, it follows (from Bestvina's work) that $G$ has finite cohomological dimension over ${\mathbb Q}$ and, if you consider torsion-free groups, over ${\mathbb Z}$ as well. (This immediately excludes Thompson's group, etc.) In particular, geometric dimension of $G$ is finite, $G$ has finite type, etc. From this you can make pretty much the same algebraic conclusions about $G$ as in the case when $G$ acts geometrically on a finite-dimensional $CAT(0)$ space. Thus, in the torsion-free case, I do not think you are missing (or gaining) much by restricting to finite-dimensional $CAT(0)$ spaces. (For instance, I do not see how assuming finite dimension of $X$ would help with proving that $G$ has finite asymptotic dimension.)
I am not sure what happens in the case of groups with torsion: It is conjectured by Swenson that a $CAT(0)$ group $G$ cannot contain infinite torsion subgroups. Maybe it would be easier to exclude some infinite torsion subgroups (say, the infinite permutation group) using the assumption that $G$ acts geometrically on a finite-dimensional $CAT(0)$ space, but I do not see how.
Swenson's work had a follow-up paper by Geoghegan and Ontaneda http://arxiv.org/abs/math/0407506 where they weaken some of his assumptions and strengthen some of his conclusions.
Note: In view of Swenson's result it is tempting to say: Take the closed convex hull (in $X$) of the ideal boundary of the $CAT(0)$ space $X$ and show that it is finite-dimensional. It might work, but, in general, convex hulls in $CAT(0)$ spaces tend to be much bigger than expected.
I think you will miss the notorious Thompson's group $F$: it acts properly isometrically on a $CAT(0)$ cube complex (see Dan Farley, http://www.users.muohio.edu/farleyds/Far1.pdf); but it cannot act properly isometrically on a finite-dimensional complex, as this would make it of finite cohomological dimension.
EDIT: Indeed as Mark pointed out, the action of $F$ on this $CAT(0)$-cube complex is NOT co-compact, hence my answer should be discarded. Don't trust MO too much.
