I think that's wrong on a couple counts. First, the main concern with this type of collection based on search terms is that it has a high likelihood of picking up users who are totally innocent. If many of those users are obscured by VPNs, that raises the bar considerably for law enforcement to de-anonymize each of them (most, if not all, of whom LE knows will be total dead ends).

Second, concerning a VPN being tapped by US law enforcement agencies, I think depending on the country in which it's being operated, that is more or less likely. These companies' entire businesses rely on the perception of security, so government interception of their communications would be an existential threat to them. I think it's also likely that interception (and crucially, use of intercepted materials as evidence in criminal prosecution) would probably get out sooner or later, just as it has in the above story. The idea that federal law enforcement has access to everyone's communications everywhere all the time probably overstates their capabilities.

As I understand it, there still remains a pretty big difference between the government's active/targeted vs passive/dragnet surveillance. A VPN is probably still useful for defending against the latter.