For comparison, a dedicated AWS m4.4xlarge instance, which is equivalent to one of these nodes without the expensive Infiniband, is about $500/month. For an actual supercomputer node you can figure about double that.
So the total running cost (of just the hardware) for global weather prediction per month is probably around 30 million USD. Then you also have to pay people to manage the running of simulations, do the analysis, produce nice plots and animations for TV, etc.
Also, I Am Not A Weather Forecast Scientist, just working in a related field (fluid dynamics). So I could be wrong about stuff here.
As for shared models and data: I know Norway and the UK do a lot of this. I imagine so does the US and Canada/Mexico.
Also, keep in mind that weather modelling relies heavily on running an ensemble with random variations in initial/boundary conditions. So a lot of the computational cost comes from running hundreds of these simulations. My guess is that you don't run the large area 100 times, just a few times and then perturb it for all your local area simulations.
In our case, we have 1000-2000 instances (the number is elastic, because AWS) handling real-time ad auctions and related data processing. Since we are just one of the many players in the industry, and far smaller than the largest ones (GOOG, FB), I am pretty sure in aggregate more computing cycles are used to optimize ads than oil or weather.
Crazy.