Despite previously stating that selling access to its Llama AI models is not part of its business strategy, Meta does make money from them. A newly unredacted court filing has revealed that the company earns revenue through agreements with businesses that host its AI models.
The filing is part of the ongoing copyright lawsuit Kadrey v. Meta, in which the company is accused of using pirated e-books to train its Llama models. It states that Meta receives a percentage of the revenue from companies that host Llama and offer it to their users.
Which companies are involved?
The document does not specify which host companies pay Meta under these agreements. However, Meta has named several partners that provide hosting services for Llama AI, including AWS, Nvidia, Databricks, Groq, Dell, Azure, Google Cloud, and Snowflake. These companies offer tools and services that make it easier for developers to run Llama models.
Although Meta allows developers to download, fine-tune, and run Llama models independently, many access them through third-party hosts for added convenience.
This latest revelation comes after Meta CEO Mark Zuckerberg mentioned the possibility of licensing Llama AI models during an earnings call in April 2024. He suggested companies reselling AI services based on Llama should share their revenue with Meta.
โIf youโre someone like Microsoft, Amazon, or Google, and youโre reselling these services, we believe we should receive a share of that revenue,โ Zuckerberg said. โThese are the deals we intend to make, and weโve already begun.โ
Metaโs AI investments and legal challenges
Zuckerberg has maintained that most of Llamaโs value comes from improvements made by the wider AI research community. He insists that an open approach to AI development benefits Metaโs products, including its AI assistant, Meta AI.
โI think itโs good business for us to do this openly,โ Zuckerberg said during Metaโs Q3 2024 earnings call. โIt helps improve our products rather than us developing models in isolation.โ
However, the revenue-sharing aspect of Metaโs AI strategy is particularly significant given the Kadrey v. Meta lawsuit. Plaintiffs claim Meta used pirated works to train Llama and facilitated copyright infringement by โseedingโ or distributing these works. They allege that Meta used torrenting methods to acquire the e-books and, in the process, shared them with others.
Amid these legal battles, Meta is pushing forward with its AI investments. The company has announced plans to double its capital expenditures in 2025, increasing spending to between US$60 billion and US$80 billion. Most of this investment will expand its data centers and AI development teams.
To help cover these costs, Meta is reportedly considering launching a subscription service for Meta AI. While details remain unclear, the paid version could introduce additional features for users, potentially marking another step in Metaโs AI monetisation strategy.
