Roll-your-own, or buy Best-of-Breed?

lets suppose you are a software vendor and you've identified a niche in the market place for a new PC based application. Lets suppose then that you get some money together and some programmers, and set about creating a plan for the build. Maybe it looks something like this:
1. Get some money
2. Get some programmers
3. Sketch out your vision for the new software application
4. Reject the Windows operating system
5. Spend all the investor money on writing a new operating system
6. Spend what is leftover on end-user functionality

Sounds like a killer project plan, right? Many people would agree. In fact, in the VR space we see any number of providers doing exactly this. But the plan is slightly different:
1. Get some money
2. Get some programmers
3. Sketch out your vision for the new software application
4. Reject the best-of-breed gaming engines
5. Spend all the investor money on writing a new graphics/animation engine
6. Spend what is leftover on end-user functionality

Hmmm, maybe I will open an online store, and instead of using PayPal, I will try and raise $5m in VC money to write a new merchant payment system.

I was CTO for enough years to recognize the importance of spending money in non-glamorous ways. Without a solid infrastructure, the flashy stuff on the desktop won't operate properly. But, I also kept as my mantra: build vs buy? in my experience, generally the people who build, have too much money provided by people who don't really understand the alternatives. These same folks are quite often the ones who like to create fiefdoms of developers. IT headcount = power, right? in the high-tech software marketplace, it is no different.

I am open to debate. Please let me know why its a better strategy to roll-your-own VR engine as opposed to buying the best available SDK.