From: https://twitter.com/itsren/status/1040313433978286082 - rendered real-time using Unreal Engine
There's no reason bar processing demands that this couldn't be done for any arbitrary 3D GIS dataset - this example seems to be just a very small area constructed specifically for the video, whereas it would be possible to zoom into any area and view the effects. Not to take away from this video, because it is certainly compelling.
Or will it be more like "If 6 feet of water is in this (fairly localized) area, this is how the local region will be affected"?
You could cut some corners by starting from the national floodplain data, which has a lot of engineering hours already in estimating the floodable area for a region, based on water height (and is what your area's flood insurance rates are based on).
The limiting factor (as another poster has commented) is likely to be the resolution of the data available, as processing power is generally good enough these days to handle any reasonable scenario - the study area is limited to the area of the storm and adjacent areas, and there are few areas where there will be a comprehensive DEM available to e.g. sub-meter accuracy. With increasing availability of drone LIDAR data we may well get comprehensive data that pushes the boundaries of what we can do, and which will require generalisation to coarser resolution. I suppose when the effects of even a few inches of water can be so catastrophic in terms of insurance claims increased resolution could be extremely valuable. For example I know that insurance companies already have sophisticated models of risk where flood risk is just one of many layers of modelled risk that they combine to avoid having too much risk in any one area.
The second type of modelling is the 3D modelling seen in the video. With a good 3D GIS dataset as seen in e.g. Google Earth you could take the results of the analysis described above and use this to determine which areas will be affected by flooding. It would then be a case of adding a layer powered by fluid simulation with the other layers in the dataset set up as 3D entities around which the fluid must flow. The point is that once you had your data set up as well as your fluid modelling, you could zoom to any location and view the results in that location, and do this on the fly - whereas in the video it is almost certainly a one-off canned job for a single location.