From the article:

"Like other players in the edge inference market, Movidius is looking to promote and capitalize on the need for low-power but capable inference processors for stand-alone devices."

Sounds like this is targeted at inference only, given the specs, that seems like a reasonable conclusion, right?

100 GFLOPS (FP16) at 1W power.

For comparison:

Tesla P40 = 11,758 GFLOPS at 250 W ($5,600).

GeForce GTX 1080 = 8,873 GFLOPS at 180W ($550).

So, a 1080 is 50GFLOPs/watt and 16GFLOPs/$ vs 100/watt and 1.27/$ for the stick.

The stick may be still useful for embedded/in the field/robotics apps?