The biggest mistake you could make at this point is locking yourself into a single network design which becomes harder down the track to scale up or down. Even a single server could technically be classified as a global presence as it can be connected to from anywhere in the world. The best option is to design your infrastructure so that it can scale up and down adding new regions and new servers within those regions without needing to take your services offline for a period of time.
I recently implemented something like this for a company that ran a managed CRM solution globally. The network started off in a single region with servers located behind a load balancer within that region. As the customer base increased and we could see the geographic areas that clients where connecting from we slowly scaled up adding services in new regions with each region having its own application load balancer and using geo-routed DNS queries to route the connection to the nearest region. At the present time the network has nearly 200 server instances behind 5 load balancer in 3 seperate regions with a customer base measured in the tens of thousands of clients and hundreds of thousands of connections. Much of the decision as to region distribution and server count will depend on how decoupled your application is, how easily it scales horizontally, and where your clients will be connecting from the most.
Best rule of thumb these days is to start off small but plan for aggressive scaling over a short period of time and to design with horizontal scaling in mind from the beginning.