The concept of a data moat is changing. Andreessen Horowitz noted in 2019 that moats erode as data grows. With LLMs now capable of ingesting any data, simply owning proprietary information is no longer enough. The new competitive edge lies in data activation—the race to make proprietary data useful and digestible for AI systems before rivals replicate the insights.

Healthcare exemplifies this shift. OpenAI and Anthropic recently launched specialized healthcare AI tools, with OpenAI reporting over 40 million daily users seeking health guidance. Yet the domain remains fragmented, exposing the inadequacy of general-purpose models. The opportunity is immense, but turning vast medical datasets into practical AI capability requires sophisticated transformation, not just raw ingestion.

Recent research shows a path forward. Studies like Tables2Traces and EHR-R1 convert structured patient data into reasoning traces that significantly improve LLM performance. These methods create contrastive comparisons or knowledge graphs to mimic clinical thinking, boosting accuracy on medical benchmarks. However, synthetic traces remain unverified and sometimes unfaithful to actual decisions, revealing the challenge of building the right “turbine” to harness data’s potential energy.