The best things to open source in this scenario are the things that you would buy from a third party, if you trusted the supplier enough. For proprietary software, a second source is almost always impossible. For hardware, it's often quite difficult, depending on the component. Switching from Intel to AMD is quite easy in a lot of cases, switching from a Qualcomm SoC to a Samsung one is more effort. Switching other components can be very hard. Service companies are a lot easier (switching from one law or accounting firm to another is much easier than retooling a production line).
Apple's involvement with LLVM is quite a good example here. Their ecosystem absolutely depends on high-quality compilers existing for OS X and iOS. With Classic MacOS and early versions of OS X, they outsourced this to Metrowerks, who produced quite a competent IDE and set of tools. Then Metrowerks, their sole supplier, was bought by Freescale and development on the Mac versions basically disappeared. They had some involvement in GCC development inherited from NeXT, but GCC was problematic for IDE integration (the parser is designed in such a way that it's impossible to use for syntax highlighting, for example - it does constant folding very early so you can't differentiate 4 and 2+2 in the source). They decided that they needed to bring compiler development in-house, but it was a lot cheaper to do so as part of an open source ecosystem. Apple now contributes something like 40% of the code to LLVM and that vast majority of what other people do directly benefits them, so they're effectively halving their costs. And, of course, giving away the IDE and compiler tools for free (rather than charging, as Metrowerks did) makes people more likely to start developing for Apple platforms.