This is following a trend. At both the AWS RE:Invent and Microsoft Ignite, Snowflake DB played a prominent role.
Similarly we have Vertica runing in EON Mode.
The local caching strategies on the compute nodes makes or breaks these sorts of systems. The separation of compute from storage means that you could have two or more completely separate data marts, each with very different load and access patterns accessing the same data. At the same time you can have a third "data warehouse" in Snowflake terms or "sub cluster" in Vertica terms that is used for ETL work. AWS S3 storage is eleven 9s resilient so that is a step up.
As the compute nodes are focussed on compute and don't have to worry about storage you can see a noticeable performance improvement. Again, I stress that it all depends on the way in which the data is cached on the compute nodes.