Most all of that is overall poor architecture, and most companies don't hire particularly good developers or DBAs (and most web developers aren't actually very good at manipulating data, relational or not), but it's the state of the union. That's "enterprise IT". That's why consultancies makes billions fighting fires and fixing things that shouldn't be problems in the first place.
A Lucene index can be much larger than your current RAM. It can be 100x that. The data will still queryable. Lucene reads into memory the data it needs in order to produce a sane result. Lucene is pretty close to being industry standard for information retrieval.
My definition is instead "when your data is not queryable using standard measures".
I unsubscribed from that (non-tech) podcast.
Since when is 10M records is considered big data?
My goto gauge for big data is that it can't fit in memory on a single machine. And since that means multiple TB[1] these days, most people don't really have big data.
[1]: *Heck you can even rent ~2TB for $14/hour! https://aws.amazon.com/ec2/instance-types/x1/