ASIC
Dedicated Vector Silicon Opens a Cost Test for RAG-Heavy SaaS
Dnotitia has brought its Vector Data Processing Unit to server scale, a dedicated ASIC for the retrieval step in RAG and agentic AI rather than for the model itself. For SaaS companies whose inference bill is dominated by search rather than generation, it is worth knowing that retrieval is becoming its own hardware category, and worth knowing that every figure published so far comes from the vendor.