If you want synchronous commit, yes, there a restriction, which is usually given as 100 km. The reason for this is that the transaction cannot be committed until the log has been hardened on the secondary replica. And the longer the distance, the longer the latency. If you wonder that defines this limit, it is simply the speed of light, 3E8 m/s.
For asynchronous commit, you could have setup like the above, but I can't say that it is recommendable. Rather, you would have one AG in Mumbai and one AG in Atlanta, and then you would join these two AG into a distributed AG.
And overall. since you normally have a Windows cluster for your AG, you probably want the cluster nodes to be close to each other for low latency.