Non-zero time, but not a lot either. Main hangup would be acquiring data for training, as their engineers would remember the parameters for GPT-4 and Microsoft would provide the GPUs. But Microsoft with access to Bing and all its other services ought to be able to help here too.
Amateurs on hugging face are able to match OpenAI in impressively short time. The actual former-OpenAI engineers with unlimited budget ought to be able to do as good or better.
