They already have a good-enough Rosetta-like x64 -> ARM translation layer. You can run any x64 windows program on ARM windows. They can optimize the performance

They already have a good-enough Rosetta-like x64 -> ARM translation layer. You can run any x64 windows program on ARM windows. They can optimize the performance more, but that’s not a whole-company effort, a small dedicated team can make the optimization happen. If Qualcomm or Nvidia start shipping M3-comparable chips I can see it going pretty smooth without large changes at MSFT.