One of the related papers I stumbled across, while not
the SHARD paper, does go into a fair amount of detail about SHARD and the problem they were trying to address. One bit of verbiage here might be illuminating:
The new SHARD) (System for Highly Available Replicated Data) system under development at Computer Corporation of America (CCA) is designed to address the problems described above. It provides highly available distributed data processing in the face of communication failures (including network partitions). It does not guarantee serializability, nor does it preserve integrity constraints, but it does guarantee many practical and interesting properties of the database.
The reader is referred to [SBKJ for a detailed description of the architecture of the SHARI) system. Briefly the main ideas are as follows. The network consists of a collection of nodes, each of which has a copy of the complete database. (Full replication is a simplifying assumption we have used for our initial prototype, many of our ideas seem extendible to the case of partial replication, but this extension remains to be made.) Replication allows transactions to be processed locally, thus reducing communication costs and delays, and providing high availability.
So it sounds to me like their main concern was availability through replication, and not so much horizontal scalability (which seems to be more the "point" of modern day "sharding"). Yet I would probably claim that there is enough conceptual overlap to say that SHARD does relate to the modern use of sharding in some sense. Although it's hard to be sure without that original paper.