A new court filing has revealed that Meta agreed with the hosts of its Llama AI model to share revenue. The filing was submitted by the attorneys of the plaintiff in the company’s ongoing Kadrey v. Meta copyright lawsuit where it was accused of using pirated books to train its artificial intelligence models.

The attorneys of the plaintiff have gone a step further to show that Meta shares a percentage of its revenue with the companies hosting its Llama models. The filing also revealed that the company shares a percentage from the funds that the hosting platforms generate from the users of the models.

This revelation is coming off the back of a blog post in July 2024, where Meta CEO Mark Zuckerberg mentioned that the company is not looking at selling access to Meta’s openly available Llama AI models as a business model. However, that is not the case as the company does make some money from Llama using a revenue-sharing formulary, at least according to the unredacted copy of the court filing.

According to reports, the filing did not discuss specifically the hosts that have the revenue-sharing model with Meta, but the company has a number of high-profile host partners. They include AWS, Databricks, Groq, Dell, Azure, Google Cloud, Nvidia, IBM, Intel, AMD, and Snowflake. Developers are not required to use the Llama models through a host partner, but these models can be downloaded, fine-tuned, and run on a range of different hardwares.

However, the hosts provide additional services and tools that ensure that the Llama models are simpler and easier while working. During a previous earnings call in April 2024, Zuckerberg had pitched the possibility of licensing access to Llama models, adding that the company could also monetize the model in other ways. He mentioned avenues like business messaging services and ads in AI interactions but failed to give out any clear specifics.

In a recent report, Mark Zuckerberg mentioned that most of the value Meta gets from Llama comes from the improvements that are done on the model from the AI research community. Meta uses the models to also power many products across its platforms and properties, including the Meta AI assistant, Meta AI.

“If you’re someone like Microsoft or Amazon or Google and you’re going to be reselling these services, that’s something that we think we should get some portion of the revenue for,” Zuckerberg said. He added that these are the types of deals that the company is now focused on making, adding that steps are being taken in that direction already.

In its Q3 2024 earnings call, Zuckerberg also mentioned that he thinks it’s good business if the company goes about it openly. “It makes our products better rather than if we were just on an island building a model that no one was kind of standardizing around in the industry,” he added.

In the argument of the attorney representing the plaintiffs, the fact that Meta may generate revenue in a rather direct way from its Llama models is important because it shows that the firm also facilitated infringement by seeding or uploading their works. The plaintiffs argue that Meta used hidden methods to obtain the e-books in a torrent fashion for training. They also claimed that due to the way torrenting works, it shared the e-books with other torrenters.

Meta has been planning to increase its capital expenditure this year, thanks in part to the increase in AI investments. In January, the company announced that it was planning to spend about $60 to $80 billion on CapEx in 2025, doubling the amount that it spent on CapEx in 2024. The funds will primarily go to data centers and growing the AI development team of the company. There are also reports that the company is considering launching a subscription service for Meta AI that will add more capabilities that are still yet to be specified to the assistant.

Cryptopolitan Academy: Want to grow your money in 2025? Learn how to do it with DeFi in our upcoming webclass. Save Your Spot