5/7/2026
Researchers from the Grainger College of Engineering's Department of Industrial and Enterprise Systems Engineering found that traditional pricing models fail in data markets because datasets can be sold repeatedly without being depleted, making stable pricing difficult to achieve. Their ICML 2026 Spotlight paper shows that allowing prices to vary based on purchase quantity can restore market stability, offering a practical framework for the rapidly growing AI and data economy.
Written by
Classical pricing theory breaks down when the commodity is data — but a more flexible pricing model restores order, says Illinois research featured in ICML 2026 Spotlight
Every time a machine learning company buys a dataset to train its models, a basic economic question arises: who sets the price, and will those prices ever settle? For most goods, economists have a well-tested answer — competition among sellers drives prices toward a stable equilibrium. It works for oil, grain and airline seats. It does not, it turns out, work for data.
Bhaskar Ray Chaudhury, Jugal Garg, and their graduate students, Eklavya Sharma, and Jiaxin Song, from the Department of Industrial and Enterprise Systems Engineering at the University of Illinois Urbana-Champaign have now shown precisely why — and identified a path forward.
The Problem with Data
Most goods are rivalrous: if you eat the sandwich, I cannot. Data is different. A dataset sold to one ML company can be sold, unchanged, to a hundred others. That single property — non-rivalry — turns out to undermine the key foundational guarantees.
Their recent paper, “Equilibrium Pricing in Oligopolistic Data Markets,” shows that when competing data sellers each try to set prices to maximize their own revenue, stable prices need not emerge at all. The most natural approach — simple, uniform per-unit pricing, the kind easy to explain and audit — is provably unable to produce a stable market. No matter how sellers adjust, there is always an incentive to deviate.
“The non-rivalry of data fundamentally alters the picture that classical theory gave us. Stability that we take for granted in physical goods markets simply cannot be assumed when the commodity is a dataset.”
The Surprising Fix
The counter-intuitive finding is that moving to a more general class of pricing — where the per-unit price can vary with the quantity purchased, rather than being a single flat rate — is enough to restore a measure of stability. Under this richer pricing model, a stable equilibrium is guaranteed to exist. Simulations confirm that sellers converge to it quickly in practice, and that real-world outcomes tend to be even better than the theoretical guarantee suggests.
Why It Matters Now
Data marketplaces — where hospitals license patient records, satellite operators sell imagery and platforms wholesale behavioral signals to AI developers — are growing as companies race to build the next generation of models. Designing these markets to produce stable, efficient outcomes is no longer a theoretical concern. This work gives market designers a concrete foundation to build from, and gives regulators a clearer lens for evaluating whether observed pricing practices are structurally sound.
The paper received a spotlight designation at 2026 International Conference on Machine Learning (ICML), a premier conference in Machine Learning, placing it among the top 2.2% of roughly 24,000 submissions.
Jugal Garg is an Associate Professor in the Department of Industrial and Enterprise Systems Engineering in the Grainger College of Engineering at the University of Illinois Urbana-Champaign. He is also affiliated with The Siebel School of Computing and Data Science. His research lies at the intersection of algorithms, economics, and game theory, with a focus on computational aspects of markets, fair division and resource allocation.
Bhaskar Ray Chaudhury is an Assistant Professor in the Department of Industrial and Enterprise Systems Engineering at the University of Illinois Urbana-Champaign. He is also affiliated with The Siebel School of Computing and Data Science. His research focuses on algorithmic game theory, data economics and the computational foundations of fair and efficient resource allocation.