This week Google announced that it has now passed 1bn cumulative activations. This is what that trend looks like.
Google generally releases these numbers at scheduled events, and they're pretty round, so we can't take them as exact. My model makes some allowances for the overall trend rather than trying to hit the exact number on the exact day of the announcement (since the figure might actually have been passed a week earlier anyway).
If we aggregate this by quarter and split out tablets (also provided by Google recently), we get this chart, showing rapid but slowing growth and tablets as a pretty small share of the total.
This base, as everyone knows, is somewhat fragmented, with a wide mix of versions in use.
The interesting thing is that if you apply these percentages to the active base (on the assumption that trailing 24m activations are active), it appears that there is very little upgrading of existing devices: the growth of new versions of the OS seems to come solely from new devices.
Of course, what's missing from this data are the devices that are used in China and do not have any Google services on them, and hence do not activate with Google. There is, obviously, no good data on this, but most people in the Chinese market think that only, say, 20% of Chinese Android devices have Google on them. That means the overall Android market is rather bigger than 1bn, with well over 200m extra devices in China, and run-rate 'unactivated' Android making up perhaps a quarter of total Android sales.
Meanwhile, as Ars Technica points out, Google is starting to get around fragmentation by putting all its services into a software layer on top of Android that it can update remotely. This solves the fragmentation problem for Google (their services are on every Android device outside Amazon and China), but doesn't help Android developers trying to work out why their app crashes on one Samsung model but not another.