Windows RT has a fatal flaw; it depends on ARM and Microsoft is acutely aware of it. As Steven Sinofsky wrote in February 2012, "End-users are technically restricted from installing a different OS (or OS version) on a device or extending the OS, so this is generally not possible, and rarely supported by the device maker".
This means that over time, hardware fragmentation is likely to happen. Windows RT started life with only three partners - Qualcomm, Nvidia and TI - each of which implements ARM differently, meaning that Microsoft has had to come up with various flavours of Windows RT for each of the ARM licensees.
This is likely to become even worse when others join the fray. Clearly this can do Windows RT no good in the long run. Enter MIPS. MIPS already is 64-bit ready and Microsoft even used to run Windows NT on it. What's more, unlike ARM, MIPS has an extensive history in high performance computing, workstations (SGI Octane anyone?), all the way down to set top boxes and routers. Plus, in theory, hardware fragmentation is less of an issue in the MIPS Ecosystem.
What's more, Microsoft happens to have a little something for the platform. For example, researchers at Microsoft have proposed a combined multi-core architecture with the capability for instruction set extension called Multicore eMIPS that runs on MIPS.
MIPS Technologies is cheap, almost pocket change for Microsoft, at around $333 million (compared to nearly $13 billion for ARM). Microsoft could buy MIPS while continuing to work with ARM and Intel, open source MIPS IP or cut its licensing fees to almost nothing and essentially do what Apple has done.
Own the hardware and own the software as well. ARM is no Intel but like Apple, Microsoft may want to own the whole ecosystem including the IP. With MIPS, it has a unique opportunity to do so.