It's practically impossible to spin stuff out of Google because of interdependencies with the rest of google3 (the monorepo), not to mention assumptions about infrastructure. You might as well completely rebuild the entire thing from scratch.

The infrastructure problem is funny because this wouldn't be a problem if Google would use GCP by themselves. They should also be smart enough to create an export function for google3.

There's a fun problem with this.

No other company in the world has workloads that match Google's. So there are two options. You can spend a shitload of money making GCP actually work for the ridiculous needs of google3 applications and then actually get everybody to use GCP rather that using borg directly or you can not spend that money building all the infrastructure to do things that borg already does and has zero external customers.

Turns out the latter is attractive.

Will there ever be a google4?

No. The google3 path itself is baked into so many things that it would be a monumental task to move. There was an attempt in maybe 2010 or so but it was quickly abandoned. Things are 1000x worse now.

What Google does now is just not rev the version in any repo paths. This is true of several projects that previously had versions included in the path. Now all versioning is handled separately.

There is one case: Google acquired Keyhole (Google Earth), John Hanke then built Ingress/Niantic which is a spunoff company (where Google afaik still holds some shares) doing quite well with PokemonGo it seems (while they had to stop their Catan project and killed off their Harry Potter game)

A couple of others: Boston Dynamics. Motorola.

Boston Dynamics was never integrated into Google, so I guess relatively easy to split of.

Motorola is interesting, but considering it was less than a year in Google and more focussed on hardware development probably also not fully integrated and one could argue whether Google acquired it only for the patents they kept when selling to Lenovo.

John Hanke in contrast built Maps and related things after being acquired and only after a few years moved to build Niantic, where at least Ingress was developed fully inside Google (but apparently not fully in Google stack, but so that they could transition to Google Cloud)

But yeah, good examples!

motorola was patents ($8B in protection IIRC) + key talent acquisition (called ATG at the time). I don't think Google ever wanted the bulk of motorola's operations or rf engineering.

They bought it, stripped out the things they wanted to keep, then spun it back out.

Google acquired HTC mobile development team to develop Pixel, after they sold Motorola. It's weird.

Selling it off incurs the risk of strengthening a potential competitor. Also, in many cases at least half of the value is in the team of developers, which probably don’t want leave Google and which conversely Google doesn’t want to lose.

It also signals to the would-be customers of their other products that the service they want to sign for has a chance of surviving. At this stage I wouldn't touch any new Google service.

Dependencies on google-only stack plus many of the things they build aren’t actually financially viable.