Question to physics replication experts

I started making an online game where the player controls a physical object with the help of forces. I know a little about the optimization of the network, so I am sure that it is not necessary to send 100 times a second information on the position of the object in the world to all the customers. All the more server have us will be one of players (host). I’m going to send position data ~10 times per second to clients, and those in turn interpolate the position from the old to the new, thus getting a move without jerks.
But the question:
As I said, the server we will be one of the players. I do not think that a player with a weak laptop will be able to calculate the physics of 16 objects in the game (the maximum number of players on the server).
I’d like to know what I should do.:
calculate all physics on the server and send replicated data to all clients,
or to carry out miscalculations on the client, sending RPC to the server, and that in turn sends this data to all other clients?

The second option seems more suitable, and the movement of the controlled object will be more smooth, but the fact is that in the future we want to release this game on mobile platforms with a dedicated server and I’m not sure that the smartphone is able to cope with this task.

If you use the second option with sending RPC, then do it Reliable or unreliable?