Yes, it's pricey, especially considering the extra components not included in the box that have to be purchased separately. If price is not a critical factor though, it undoubtedly has advantages for projects where a high degree of custom setups, compact form factor or peripheral attachments are required.
Maybe good for development while trying to figure out what gizmo and horse power you need.
If you are using Deep Learning I am not sure the GPU gives you any real advantage because it doesn't hardware
accelerate Googles Tensor Flow. I find it odd that AMD isn't doing that? It seems that is staying away from AI
by not offering GPU acceleration for Tensorflow.
The Intel Movidius has still to come up with a solid offering that you can just plug into Tensor Flow .
Some new models of the NUC 8 are coming out with a lower graphics specification, and so probably a lower price.
Hope this makes it cost effective enough to at least try out. The current crop of things have no tensorflow acceleration and the Movidius Neural stick I am finding to be extremely difficult to work with. Sure I can get their samples working at a couple at 3fps on a PI but try and reuse Inception V3 with your own classifications and you find its hard, so much so you find that your just guessing with no solution in sight.
Oh well. it it was easy...