Let me try to sketch a proof of "the evident fact that sum of angles around a vertex of a convex polytope is less than $2\pi$," as phrased by Alexandre Eremenko.
This proof below is more complex than Otis Chodosh's, but perhaps has the advantage
of using convexity explicitly.
Let $v$ be a vertex of the convex polyhedron $P$.
Intersect $P$ with a small sphere $S$ centered on $v$.
The intersection forms a convex spherical polygon $Q$ (blue below)
lying within one hemisphere of $S$.
(The hemisphere rim (red below) corresponds to a plane supporting $P$ at $v$.)
We can take the radius of $S$ to be $1$ by appropriately scaling $P$ prior to
intersecting.
Note that the angles of the faces of $P$ incident to $v$ become the arcs of $Q$.
Now I would like to claim that the perimeter of a convex spherical polygon $Q$
lying within a hemisphere of a unit sphere $S$ is at most $2\pi$.
I will mimic a proof in a
Noam Elkies posting for
a planar analog of the claim.
The proof is by induction.
Let $Q_0$ be the hemisphere rim; its length is $2\pi$.
$Q_0$ contains $Q$: $Q_0 \supseteq Q$.
We will form a succession of more tightly enclosing convex
spherical polygons, $Q_0 \supseteq Q_1 \supseteq Q_2 \ldots \supseteq Q$,
with decreasing perimeters (more accurately: non-increasing perimeters).
Let $k$ be the number of edges of $Q$ that do not lie along
the boundary of $Q_i$. If $k=0$, then $Q_i=Q$ and we are finished.
Let $e$ be some edge of $Q$ that does not lie on the boundary
of $Q_i$.
Extend $e$ to a great-circle arc that cuts $Q_i$ at points $a$ and $b$.
Define $Q_{i+1}$ as following $Q_i$ but then the "shortcut" $ab$.
Because $ab$ is a geodesic, it is a shortest path between $a$ and $b$,
and therefore cannot be longer than the portion of $Q_i$ cut off.
So the perimeter of $Q_{i+1}$ is at most the perimeter of $Q_i$.
Continuing in this manner produces the nested sequence of
enclosing spherical polygons, ending with $k=0$.
If no edge of $Q$ lies along the original $Q_0$, then
the first cut will be a semicircle (green below), and
the perimeter of $Q_1$ is $\pi+\pi$, i.e., still $2 \pi$.
So the perimeter of $Q$ is
at most $2\pi$, the perimeter of $Q_0$.
Therefore the sum of the angles of the faces of $P$ incident to $v$
is at most $2\pi$.
Therefore, $P$ has positive angle deficit at $v$, i.e., it has positive Gaussian
curvature there.
Note convexity is used crucially, for a nonconvex spherical polygon could have
an arbitrarily large perimeter, still fitting inside a hemisphere.
Added (
2Jul2015).
The example below is intended to show it is not correct
to project the angles incident to $v$ to a plane and argue that
the angles only enlarge under projection.
Below, one angle is $68^\circ$ incident to $v$ in 3D,
but that angle projects to $45^\circ$ in 2D.
As you can see, some angles grow, some shrink, under projection.