“Cerebras solved the AI data movement bottleneck by building a large, wafer-scale chip that co-locates memory and compute.”