New CMU, Anaxi Labs collaboration explores how AI ecosystems create and share value

A new collaboration between Carnegie Mellon University and Anaxi Labs aims to tackle one of the biggest unanswered questions in artificial intelligence: how the rapidly expanding AI ecosystem should distribute the economic value it creates.

The partnership brings together researchers from Carnegie Mellon and industry experts at Anaxi Labs to study the emerging economic foundations of generative AI systems.

This collaboration will focus on two major challenges shaping the future of AI: how to value and compensate the data that powers AI models, and how AI platforms themselves should generate revenue as LLMs and agents replace traditional search engines.

The project is rooted in research led by Chenyan Xiong, an associate professor at Carnegie Mellon’s Language Technologies Institute in the School of Computer Science. Xiong has spent several years studying how training data influences the performance of large AI models, and his work now intersects with broader economic questions about how the benefits of AI should be shared.

“AI is changing how we work and how the digital world itself is built,” said Xiong. “With the new AI-native format, where does the revenue come from, and how do we distribute it? Those are very critical questions that researchers need to help answer.”

Modern AI systems rely on enormous volumes of training data, ranging from publicly available information to curated datasets produced by human annotators. Yet the value of that data is rarely clear.

Xiong’s research explores ways to estimate how much individual data points contribute to a model’s performance. By analyzing how training data improves model accuracy, researchers can approximate what he calls the “value” of that data.

“We study the improvement of model performance when the data is used for training,” Xiong explained. “If we can estimate how much a specific dataset contributes to the model, we can begin to understand what that data is actually worth.”

Those insights can help address a growing concern in the AI industry: that the global labor force responsible for labeling and preparing data often receives little to no compensation, despite playing a crucial role in AI development. Some annotators are paid only pennies per task, while specialized experts who generate high-quality datasets may command far higher wages.

Xiong believes tying compensation to the measurable value of data could help improve both fairness and quality. In the follow up studies conducted on top of their paper “Fairshare Data Pricing via Data Valuation for Large Language Models,” with collaborators across Carnegie Mellon, including colleague Beibei Li at the Tepper School of Business and the Heinz College of Information Systems and Public Policy, researchers found that when data contributors are rewarded based on the value their work adds to AI models, they tend to produce higher-quality data.

“If workers know their bonus depends on the value of the data they produce, they have strong motivation to provide better data,” said Xiong. “That can create a win-win situation where model developers receive better training data and contributors are compensated more fairly and more per hour.”

AI is becoming less like a single product and more like an ecosystem.

Kate Shen, co-founder, Anaxi Labs

The collaboration with Anaxi Labs extends beyond data valuation to explore how generative AI platforms themselves should operate economically.

As generative AI systems increasingly provide direct answers rather than links to websites, the traditional advertising-driven model of the internet faces new pressure. Generating AI responses requires significant computing resources, and the ad revenue that historically supported online platforms often depends on user clicks: interactions that may disappear when AI summarizes information instantly.

The first joint paper released through the collaboration,“An Economic Framework for Generative Engines: Advertising or Subscription?,” examines how AI platforms might balance different revenue strategies as this shift unfolds.

According to Kate Shen, co-founder of Anaxi Labs and CMU alumna, the goal is to understand how economics design will shape the future AI ecosystem.

“AI is becoming less like a single product and more like an ecosystem,” said Shen. “Agents call other agents, models rely on curated datasets maintained by contributors, and the systems behind them need economic models that fairly evaluate and reward those capabilities.”

Anaxi Labs is developing infrastructure designed to support that kind of ecosystem. The company is building a global data supply chain for AI and robotics, and a programmable marketplace where datasets, prompts, AI agents, and workflows can be shared as reusable components and monetized. Contributors can publish their assets, track how they are used, and automatically receive a portion of the revenue generated when their work powers downstream AI applications.

The idea reflects a broader shift in the AI industry, where competitive advantage is increasingly tied not only to model architecture but also to access to high-quality data and specialized expertise.

The research team behind the CMU/Anaxi Labs collaboration is also exploring how AI could reshape the structure of the internet itself.

Today’s web relies heavily on centralized search engines that index and rank vast amounts of online content. But Xiong suggests that future AI systems may operate more like networks of interacting agents, with specialized systems retrieving information directly from content providers rather than storing everything in a single index.

That shift could create new challenges for how information is distributed and monetized online.

“As AI becomes the primary interface to knowledge, the systems behind it must reward the people who produce valuable content and data,” said Xiong. “Otherwise, value could concentrate in a very small layer of model providers while discouraging the contributions that make AI systems useful.”

For CMU and Anaxi Labs, the work is only beginning. With AI technologies expected to shape industries ranging from software and cybersecurity to healthcare and media, the economic rules governing the technology could have far-reaching consequences.

“This is a multi-trillion-dollar problem that will shape society,” said Xiong. “We’re still at the very beginning of understanding how it should work.”