They already have a terrible case of NIH syndrome. It is already well accepted at Google to build if you can avoid the buy.
Usually, I'd consider that a bad thing. However, they have proven they can write some damn good software. They have also proven that they can run a data center better than anyone else can. Both their internal and external systems are a pleasure to use. NIH is only a problem when it starts to burn you, but that hasn't really happened... yet.
Luckily, they count most open source software as a building block. If there already exists a great open source package, they aren't going to completely ignore it for the sake of NIH syndrome. It's about control, not price.
Put simply: Google demands more from their hardware and software than any other organization. Moving and sorting bits is their core competency. This seems like a smart move to me.
Another reason NIH makes sense for Google is that since they compete in so many markets, the shifting alliances and mergers and whatnot for their vendors could mean that they could end up in a situation where a competitor buys a vendor which puts their support/upgrades/pricing at risk.
It makes a lot of sense. Goog has been using Juniper for a while, so goog really understand the problem domain and what % of that problem is solved by Juniper's products.
They would also know what they like & dis-like in terms of features their product should have vs. juniper which probably address a wider customer wish list; maybe & I am speculating here, Juniper takes a long time to turn around on features or some features are custom to goog and Juniper doesn't want to build them.
So goog gets to design a product that fits their needs for the next few years. If the product turns out well there might even be tremendous competitive advantage since googs competitors can't buy it.
Worst case things don't work out and they go back to buying Juniper.
The odds that Google has the best whatever engineers in the world diminishes as whatever gets further and further from search and advertising. Are they going to be designing their own processors? Fabbing them? DEC is the last company I can think of that was fully vertically integrated, even Apple isn't.
Vint Cerf is already under Google's patronage. I am pretty sure TCP/IP is not exactly Google's "core" business! No matter how far the "whatever" is, it is clearly helping and supporting search and advertising.
For the scale at which they operate it probably makes tremendous sense.
Although they surely exert a lot of influence on their vendors, they are still beholden to those vendors to release patches and features on a timeline they cannot control.
This is, IMO, an evolution of the open-source software model. They have a great template to develop/benchmark against, they know what a router in their datacenter should cost, how it should perform, etc. They can create something with the specific features they want and have an immense understanding of a platform that is crucial to their operations. Most likely for much the same reason those chose and open-source model in the first place for their server platforms.
There has also been a fair bit of volatility it the router/telecom markets over the years. By owning core technology they are better equipped to maintain their business continuity.
