I remember asking a USGS person about this. They remarked that the other difference was that, compared with the OS, the USGS data was a bit rubbish (I may be paraphrasing).

The USGS is funded by some shard of the US federal budget, and does commendably good stuff with the budget it gets; it's there for both high-minded and commerce-supporting reasons. The OS is now (in a sequence of reorganisations from 1990 to 2015) a private company with a government-owned golden share, and is expected to be revenue-positive. The fact that it has more money per square metre of country, means that it's able to be _very_ thorough, mapping down to the level of individual bits of street furniture.

Sidenote: the context I was hearing this included a talk by someone from OS describing using reasoning software to do consistency checking of their GIS: for example, if you find a river bank in the middle of a field, something has been mislabelled. I thought that was cute.

When you buy a data product from OS, you're buying some subset of the layers of the database.

As the other reply pointed out, some of these layers are available for free, and in the last few years there's been some review/churn/debate in the data subsets made available that way (I see there are more details on the Wikipedia page). One can form a variety of opinions on whether those subsets are as big as they could or should be, but there does seem to be a substantial point that the level of the detail in the master map is there because it's profitable for the company (and thus income-generating for the government) to develop it from surveys, and it wouldn't exist otherwise.

I think the Met Office is organised in a similar way.

There are a number of questions of principle and practice here, but the OS seems to me to be claimable as an example (rare, in my opinion) of a privatisation which has produced net positive outcomes.

OS does release a large volume of open data, but yes, the vast majority of the good stuff is not open.

https://osdatahub.os.uk/downloads/open