# Novelty vs Volume in Data Value Data value scales with novelty, like IP, or with volume, like oil. The answer sets the shape of the data market. - Claim to test: small firms sell data for a large share of revenue. Very large firms sell it for a small share. - That gap suggests diseconomies of scale in data sales. - Five causes can explain it. They have different futures. ```mermaid graph TD Q[Why does large-scale data sell for less per unit?] Q --> A[Few buyers can write a large cheque] Q --> B[Distressed sales set the price] Q --> C[Extra data adds little after a point] Q --> D[Cleaning and transfer cost rises with size] Q --> E[Synthetic data replaces real data at scale] A --> T[Temporary: prices can recover] B --> T C --> S[Structural: novelty wins] E --> S D --> I[Opening for data infrastructure] ``` | Cause | Durable? | Implication | |---|---|---| | Thin buyer pool | No | Wait for the market to mature | | Distressed pricing | No | Ignore these comparables | | Diminishing returns | Yes | Sell rarity, not bytes | | Transfer cost | Partly | Build the pipe | | Synthetic substitution | Over time | Real data sets the baseline only | ## What follows - Price a dataset by what a model cannot learn elsewhere. - Rare events and expert trajectories hold value. Bulk logs lose it. - [[Sellable RL Environment]] is the novelty answer in product form. ---- tags: #boat #AIstrategy #economics links: [[Scarcity Migration]] [[Concepts MOC]] [[Data Moat]] [[Data Capital Ladder]]