Google's founders did the up-front R&D in academia: they were Stanford grad students who developed PageRank, and then launched a company to take it to market. And Microsoft's big breakthrough was based on technology they didn't even develop! They got a contract to deliver an OS for IBM, but has no OS, so they bought 86-DOS, quickly made about 2 months of modifications to get it to fit the specs IBM demanded, renamed it to MS-DOS, and delivered it to IBM. Or take Sun: they shipped a computer architecture that Andy Bechtolsheim had largely developed while a grad student at Stanford, paired with a commercialization of UC Berkeley's 4.1BSD (rebranded SunOS 1.0).
I think startups can indeed be a good way to take existing R&D and productize it, i.e. "make it out of the lab", but I don't think they develop a lot of it.
The Google guys' PhD work was a prologue to their startup. They went into it with novel technology that they had developed themselves. There is a huge difference between this approach and saying "no more R&D, lets take what's out there and build a product with it." When they started Google they certainly did plenty of R&D to improve their algorithms and crawler, and it's been continuing ever since.
For Microsoft the reference there was not their (extremely lucky) deal with MS-DOS but all of the other work they did: from the early BASIC days to the days of Windows where they did extensive R&D in order to leapfrog their competitors.
You ignored my other examples but the fact is that the Lean Startup approach to startups is a relatively new thing, only really fits for web-oriented startups where up front development costs are low thanks to SAAS and open source, and even then the companies you might be able to point to as definitive success stories are YC companies and maybe companies that are the latest round of tech IPOs taking a Lean slant on how they found their product market fit. Eric Reis's startup failed! Find me a biotech startup, find me an energy startup, find me a robotics or hardware startup that doesn't do R&D on day zero. The YC model of developing a web application using open source software and launching on Heroku to get revenue for a B2B need is a very narrow slice both in time and space of the way innovation happens. Hell, look at any legendarily successful software companies that aren't in the web sphere of the last 10 years, like iD Software, Adobe, Pixar, Wolfram, and so on and you see R&D focused companies from their beginning.
I'm not really defending the "Lean Startup" approach; I don't think it's very socially useful or interesting, but I do think it's probably profitable on average (hence why VCs like it).
The approach of doing a university spinoff using your own tech is a model I do like, and I agree it's a way to develop new tech and also bring it out of the lab. But I think the academic part there is key: you get some R&D runway up front before you start the startup and need to make money, rather than doing a startup up-front. You can then "afford" to start the startup once at least some of the high-risk work has been done and you're reasonably close to a product. Since you mention Wolfram, that's probably an even better example of that than Google, given Wolfram's rather lengthy prologue to his startup: he spent 8 years at Caltech (1979-1987, first as PhD student, then as professor) developing the Symbolic Manipulation Program, a precursor to Mathematica. Then he spun off a company in 1987 and launched Mathematica in 1988. I don't think he would've been able to do the same kind of initial R&D in a startup as he was able to do at Caltech.
fair enough -- i misunderstood your post. i consider "PhD turns their research into a startup" to be a great approach and completely antithetical to the lean startup model which says you do customer development first, not putz around in your lab for 5 years developing nascent tech.
But when would you decide to transition from research to startup? Probably when you decide that you can apply that research to something profitable, and do it very quickly, and when the research seems to be at a point where development is scalable (i.e. if you throw more people at the problem then the problem will be solved faster.)
I'd think you'd want most or all of the experimenting to be out of the way.
I think startups can indeed be a good way to take existing R&D and productize it, i.e. "make it out of the lab", but I don't think they develop a lot of it.