For now, it is all speculation and rumour, even if it turns out to be accurate (or not). The only thing we really have 'as Ampere fact' at this stage are the official numbers from the Ampere A100 ('enterprise' model), also per Anandtech below. While the general Ampere architecture will be applied to consumer and pro_sumer cards, there will obviously be different configurations of memory (HBM2 vs GDDR6, size), and of the number of Cuda, Tensor and RT cores. The A100 is (currently at least) the flagship, with 6912 Cuda cores...earlier rumour / speculation had the full-fat GPU at 8192 Cudas, but 6912 is it, for now at least (also depending on yields, like in some previous gens where the full theoretical size was never released).
What caught my attention is the power consumption (400w @ ~ 1.41 GHz) of the 7nm A100, a lot higher than previous (12nm) gen, though it is certainly more powerful. The A100 has 54 billion transistors and is physically still humongous, even if it is 7nm. It does suggest though that the consumer and pro_sumer 7nm models will likely be quite 'toasty' and will probably benefit greatly from water-cooling. There is the related question of die shrinks focusing heat on smaller areas that are harder to cool. Finally, it will be really interesting to see how the 7nm wafer yields shape up as that will also have an impact on pricing. Personally, I couldn't be happier with my two 2080 Tis for now...but when the time comes, I hope to see some nice custom PCB factory w-cooled models on the market, perhaps in early 2021.