It is to collect more data for LLM training. To tap into proprietary code bases that are never uploaded to GitHub.
Even if they explicitly and clearly state that they don't use your data for any purpose other than generating immediate responses, they can change this once the free Copilot gains traction and people become really addicted to it.
Like, "pay if you don't want us to use your data for training". Most people won't pay and will be happy to give away their data instead.