Actually more like 2, as Cython isn't a Python implementation. So, they're talking about Tauthon and PyPy.
I'm not so sure about Tauthon -- There hasn't been much activity since the original work, even though a new 3.x version is out which presumably has features that could be backported (pathlib.Path().resolve() now has a strict parameter which, when False, makes it act like normpath() instead of abspath() yaay!).
But then they're also forgetting a newly launched Python implementation that decided to go with 2.x even though 3.x has been out for a decade:
Nobody knows how long any Google project is going to be around, but Grumpy is alive and kicking, and I wouldn't be surprised if enough commercial users tie themselves to it to want to keep it alive.
There's a 3.x branch somewhere, but it's not in the official Mercury repo, and even though 2.7.0 was back in 2015, 2.7.1 is under active development with commits almost every week.
Correction: No, the new Path().resolve() doesn't act like normpath(). It still follows symlinks if they exist, it just tolerates non-existing paths. Bummer.
> Based on the rate of uptake for Python 3 libraries and the rate of decline for Python 2-only libraries, the Microsoft team predicted that 3 will overtake 2 by "around May of this year."
That's in number of uploaded package versions each month. Interestingly, py2+3 packages are the largest category of packages, beating both pure py2 (since May 2014) and pure py3 (since forever), and py2+3 is still growing faster than py3 too.
@vertigo I think Tauthon is the only one going for that, backporting everything possible from 3. The rest are probably going to be identical to CPython 2.7 forever. But time will tell.