Sharding is one of the most prominent concepts which involves the division of the network into shards for concurrent processing of transactions. Different sharding protocols are being implemented in blockchains to enhance its scalability. The existing blockchain systems create shards using proof-of-work consensus protocol. This research aims at developing a machine learning-based sharding process that uses the nodes’ geographical locations—latitudes and longitudes. IP addresses of the nodes are mapped to geographical coordinates, and these coordinates are then divided into shards using a suitable clustering algorithm. The nodes in the shards are geographically closer, thereby reducing the propagation delay in the network during intra-shard communication. GeoSharding has been tested to be significantly faster as compared to PoW-based sharding. This optimizes the network sharding process, thus escalating the scalability to a new level.