The Snowflakes minimum Is something my colleague got from them when we set up an account for a startup idea.
I've tried the same in snowflake, and I'm only using their smallest warehouse, and everything happens in seconds. Loading GBS of data is seconds, scanning GBs of data is seconds. Their smallest compute warehouse is still significantly beefier than the cpu you get with a puny lightsail with a shared vcpu and while it's disk io is shit (since it's reading from S3) the parallelism makes up for it.
In the end I think it's the pricing model that makes the difference. With snowflake you seamlessly launch a very powerful machine to run your query and only get billed for a minute, and with running postgres you need to run your machine all the time. Also hot storage (as opposed to s3) is also non trivially expensive. Haven't seen any way to get 200gb of ssd storage without spending that 25 dollar minimum that snowflake costs anyway. This is data I intend to not even update or query more than a few times a month, so I really don't want to pay for compute running all the time in its name.
If you drop the SSD requirement, a cheap dedicated server will blow Snowflake out of the water. As a bonus, you can run your own code (ETL, scrapers, dataviz). Kimsufi has 4 core 2TB (spinning) w/ 16GB RAM for $17/mo. Personally, that is the route I go for my personal warehouse, as I find I almost always want to put an API or django app in front (and other software like Celery, scrapers, etc.)
I'm not actually bothering to run backups because theoretically ALL of the data there can be retrieved from other sources - pulled back out of the Twitter archive exports for example.
But another similar project uses tarsnap for backups, which is pretty inexpensive.