In the article, it makes the point that Qualcomm, it seems, is choosing neither but using their own architecture. That point right there means we'll have to see the actual silicon to "know" what's under the hood. Here? I can already see a use in my mostly integer model-building/analytics. The rest of them usually needs no more than single precision. [For the latter and DP, I throw a 1-2 TFLOP machine at but that beast thinks electricity is a snack-food.]
I can already see why Google is interested. Now what's interesting are the loads that can use 24 cores out-of-the-box.