Srikanta Datta Identifies Request-To-Silicon Gap in Enterprise AI

Srikanta Datta, Director of AI at Coupang Global LLC, has identified a critical operational blind spot in enterprise AI known as the Request-To-Silicon Gap. This gap refers to the inability of organizations to trace a single business transaction through agentic loops, retrieval steps, and model calls down to the specific physical compute, memory, and energy consumed.

Datta argues that while enterprises use various monitoring tools for infrastructure and model observability, these systems lack the correlation required to determine the actual cost per request. This deficiency becomes particularly acute as AI architectures move toward complex agentic systems. Gartner estimates that one third of all enterprise AI interactions will be agentic by 2028, and in such systems, a single request can trigger numerous downstream operations.

The lack of causality and unit economics leads to unpredictable cost variances and margin erosion. Datta suggests that scalable AI operations require integrated capabilities for correlation, causality, and automated recommendations to close this gap and establish precise cost accounting.