take a look at Cerebras, who are doing wafer-scale compute