I can see variants of the Cerebras approach taking slices of the VC investment pie. For example, arrays of identical and simple chips with in-chip memory could

I can see variants of the Cerebras approach taking slices of the VC investment pie. For example, arrays of identical and simple chips with in-chip memory could have similar performance as the monolithic Cerebras wafers-scale chips. Not for all workloads, but some, such as inference.