Anaxi, Carnegie Mellon study AI data value, revenue models

Anaxi Labs has partnered with Carnegie Mellon University to research the economics of artificial intelligence, focusing on how value is created and distributed in emerging AI ecosystems.

The collaboration brings together Carnegie Mellon researchers and staff at the New York-based company to examine two areas: interactions between AI agents, and the valuation and pricing of datasets. They have also released a joint paper, "An Economic Framework for Generative Engines: Advertising or Subscription," on revenue models for AI platforms as generative systems replace traditional search and discovery tools.

Shift in economics

The project centres on a question that has grown more urgent as generative AI tools shift from link-based search to direct answers. While that change may make services easier to use, it also challenges the advertising model that has supported much of the internet economy.

The research will explore frameworks that combine advertising and subscription approaches rather than treating them as fixed alternatives. It will also examine methods for measuring how datasets contribute to model performance, with implications for how AI developers pay for the information used to train systems.

Ecosystem view

Kate Shen, Co-founder, Anaxi Labs, said AI products now depend on a growing network of contributors and systems rather than a single model acting alone.

"AI is becoming less like a single product and more like an ecosystem," said Shen. "Agents call other agents, models rely on curated datasets, and each output is increasingly the result of many contributors' work. The systems behind this shift need economic models that fairly evaluate and reward those contributions."

The academic lead on the collaboration is Chenyan Xiong, Associate Professor, Carnegie Mellon University's Language Technologies Institute. His work has focused on information retrieval and machine learning, including research into how data quality affects AI system performance.

"Modern AI systems rely on enormous volumes of training data, but the value of that data is rarely clear," said Xiong. "If we can estimate how much a dataset contributes to model performance, we can begin to understand what that data is actually worth."

Data valuation

The question of what training data is worth has become more prominent as developers face pressure over sourcing, copyright, compensation and the concentration of value among a small number of model providers. Large language models depend on vast quantities of text, images and other material, yet there is little agreement on how to price those inputs or reward the people and organisations that supply them.

Earlier research by Xiong and Beibei Li, Professor of IT and Management, found that paying contributors according to the measurable value of their data led to better AI models. The new partnership aims to build on that work by extending the analysis to broader AI systems and agent-driven environments.

This matters because the structure of AI products is changing. Instead of one model handling every task, companies are increasingly experimenting with systems in which specialised agents call other agents, combine third-party tools, and draw on curated datasets or task-specific knowledge bases. In such arrangements, the final output may reflect many separate inputs, making attribution and payment more complex.

Anaxi describes its business as AI data infrastructure and says it is building a marketplace where datasets, prompts, agents, skills and learning frameworks can be shared and monetised based on use. It argues that these components will become reusable building blocks within larger AI systems, creating a need for clearer rules on pricing and compensation.

Revenue models

The first paper from the partnership addresses another unresolved issue: how generative AI services make money when they reduce the need for users to click through to websites. Search engines and many online publishers have long relied on advertising tied to web traffic, but direct-answer systems can weaken that model by keeping users inside the AI interface.

The research examines whether AI platforms can adopt more flexible combinations of advertising and subscription pricing as these systems mature. The debate has become central for developers of chatbots, search products and AI assistants, many of which face high computing costs while users have shown mixed willingness to pay subscription fees.

Wider questions

The work also touches on a broader policy and market concern: whether the incentives built into AI systems will produce fair outcomes for contributors across the supply chain. Researchers argue that if data creators, domain experts and developers are not paid in line with the value they add, quality may fall and participation may narrow.

The collaboration is intended to explore these issues before business models become fixed across the sector. As AI tools take on a larger role in knowledge access, software use and decision-making, the economics behind them may prove as important as the underlying models themselves.