I thought this at first too, but after re-reading it became clear that they're saying IBM's process takes up the least room per transistor on average while Intel's takes up the most. So IBM is actually beating Apple by this metric here, however counter-intuitive that may seem.
IBM's 2nm beating TSMC's 3nm doesn't seem counter-intuitive. If their naming systems are comparable it's exactly what you'd expect.
Well, everyone knows now that the 'X nm' naming doesn't actually mean much when looking at the whole chip, which is why we're talking about transistors per area. Also, Apple has been designing modern chips with massive production volumes for a while, and IBM... hasn't (in fact, they've mostly sold off their former fabrication capabilities). So that's why my expectations were a little different.
That makes sense as well. It's not unexpected that a one-off F1 car is faster than a mass-produced Mustang.
No, Intel's 10nm is a bigger process, lagging behind TSMC's 5nm, so its per-transistor area is bigger.
I see, I was reading the numbers as a density (per nm^2)