My experience tells me it’s not a good idea. I guess it depends on what granularity you need for the census data. You’ll blow up the record limitation in the Pro plan if you go more granular than the county level (I have no idea about enterprise).
Secondly, as much as I like Airtable, it’s slow. For any kind of real computation, I would want to use something else. If you can separate out the processing of the data from the things you would like to use Airtable for, then push processed product to your base, I could see that working.
I use R for census data work, but I’m sure Python has similar libraries to speed the process.