They used to buy it all in, but seem to be sourcing more and more internally and algorithmically derived from public sources. This is good because it gives them more control and licencing flexibility, but has led to a (temporary?) decline in quality and some very odd errors e.g. fictitious villages based on typos and roads in the UK having french names.
I wouldn't judge OSM by US data by the way. Each country has different levels, partly based on what their Governments have released and what the community is like. Which is partly true of Google as they rely on volunteers to contribute in some nations.