Meta's Revenue Sharing with Llama AI Model Hosts Revealed

Meta's Revenue from Llama AI Models

Mark Zuckerberg, CEO of Meta, previously stated in a blog post last July that “selling access” to the openly available Llama AI models was not aligned with Meta’s core business strategy.

However, a recently unredacted court filing indicates that Meta does, in fact, generate income from Llama through revenue-sharing arrangements.

Details from the Court Filing

The filing, submitted by legal representatives for the plaintiffs in the copyright lawsuit Kadrey v. Meta, reveals that Meta receives “a percentage of the revenue” earned by companies that host its Llama models and offer access to users.

Meta is currently facing accusations in the Kadrey v. Meta case of utilizing hundreds of terabytes of illegally obtained e-books to train its Llama models.

Llama Host Partners

While the specific hosts contributing revenue to Meta remain undisclosed, the company has publicly listed several Llama host partners.

• AWS
• Nvidia
• Databricks
• Groq
• Dell
• Azure
• Google Cloud
• Snowflake

Developers have the option to download, customize, and operate Llama models on various hardware configurations without utilizing a host partner.

Nevertheless, many hosting services provide supplementary tools and services that streamline the deployment and operation of Llama models.

Zuckerberg's Statements on Monetization

During an earnings call in April, Zuckerberg discussed the potential for licensing access to Llama models and exploring other monetization avenues.

These included possibilities such as business messaging services and advertising within “AI interactions.”

Zuckerberg explained that Meta intends to receive a portion of the revenue from companies like Microsoft, Amazon, and Google that resell Llama-based services.

He stated, “So those are the deals that we intend to be making, and we’ve started doing that a little bit.”

Value from the AI Research Community

More recently, Zuckerberg emphasized that the primary benefit Meta gains from Llama stems from enhancements to the models contributed by the AI research community.

Meta leverages Llama models to enhance various products across its platforms, including its AI assistant, Meta AI.

Zuckerberg believes that an open approach is beneficial, stating, “[I]t 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.”

Implications for the Kadrey v. Meta Lawsuit

The revelation of direct revenue generation from Llama is particularly relevant to the Kadrey v. Meta lawsuit.

Plaintiffs allege that Meta not only employed illegally obtained materials to develop Llama but also facilitated copyright infringement by “seeding,” or uploading, these materials.

The plaintiffs claim Meta utilized covert torrenting methods to acquire e-books for training purposes, inadvertently sharing them with other users through the nature of torrenting.

Increased Investment in AI

Meta is planning substantial increases in capital expenditures this year, primarily driven by growing investments in AI.

The company anticipates spending $60 billion to $80 billion on CapEx in 2025, a significant increase compared to its 2024 CapEx, with a focus on data centers and expanding its AI development teams.

To help offset these costs, Meta is reportedly considering launching a subscription service for Meta AI, offering additional features to subscribers.

Updated 3/21 at 1:54 p.m.: A Meta spokesperson directed TechCrunch to an earnings call transcript for further clarification. A relevant quote from Zuckerberg regarding Meta’s intention to share revenue with major Llama model hosts has been added.