A Sustainable AI Economy Needs Data Deals That Work for Generators

TL;DR

The paper argues that the machine learning value chain is structurally unsustainable because data generators are economically marginalized by existing data deals. It analyzes 73 publicly disclosed deals to diagnose an Economic Data Processing Inequality driven by invisible provenance, asymmetric bargaining power, and inefficient price discovery. It then proposes the Equitable Data-Value Exchange (EDVEX) framework, featuring task-data matching, auditable provenance, and dynamic, utility-driven pricing to balance incentives and enable equitable revenue sharing via mechanisms like Shapley values. If realized, EDVEX could foster broader data participation, improve data quality, and create a more transparent and stable data marketplace that sustains AI development.

Abstract

We argue that the machine learning value chain is structurally unsustainable due to an economic data processing inequality: each state in the data cycle from inputs to model weights to synthetic outputs refines technical signal but strips economic equity from data generators. We show, by analyzing seventy-three public data deals, that the majority of value accrues to aggregators, with documented creator royalties rounding to zero and widespread opacity of deal terms. This is not just an economic welfare concern: as data and its derivatives become economic assets, the feedback loop that sustains current learning algorithms is at risk. We identify three structural faults - missing provenance, asymmetric bargaining power, and non-dynamic pricing - as the operational machinery of this inequality. In our analysis, we trace these problems along the machine learning value chain and propose an Equitable Data-Value Exchange (EDVEX) Framework to enable a minimal market that benefits all participants. Finally, we outline research directions where our community can make concrete contributions to data deals and contextualize our position with related and orthogonal viewpoints.

Figures (5)

• Figure 1: Recent data deals across past calendar years. See \ref{['app:data-deals-full-table']} for full list.
• Figure 2: A pipeline symbolizing a piece of the data value chain in machine learning and the structural defects underlying the economic data processing inequality. 1) Data aggregators often strip provenance information of data generators when selling data to companies that transform data into model weights and monetizable products. 2) Model monetizers, which often also transform the data to model weights, enjoy a bargaining advantage as they control much of the current revenue generation. 3) Due to their heterogeneity, data generators in particular are not well equipped to participate in the price discovery of their own data.
• Figure 3: Counts of data deals with and without revenue share information. Left bar (No revenue share): solid segment (15 deals) corresponds to deals where public sources quoted a revenue volume. If ranges are given we conservatively take the floor of that range. Total sum of disclosed revenue volume is $677.3m. The dashed segment (52 deals) indicates additional deals without revenue share and shows the problem of dark figures in this space. Right bar (Revenue share): 6 deals indicate revenue sharing with generators, only one has public information on actual revenue ($2.5k).
• Figure 4: EDVEX patches for a sustainable and efficient machine learning economy.
• Figure 5: (Top): The current landscape of AI data deals is largely dominated by transactions with large-scale content holders (such as major publishers), lacking efficient price discovery mechanisms. Data from smaller players is often scraped en masse without compensation or simply overlooked. (Bottom): Our envisioned EDVEX Framework features efficient task-data matching, utility-driven data pricing, and auditable provenance to create a more efficient, equitable and transparent ecosystem.