Meta engages in revenue-sharing for Llama AI models

Meta has revealed in a recent court filing that it earns money through revenue-sharing agreements with hosts of its Llama AI models. This information contrasts with statements made by CEO Mark Zuckerberg last July, when he claimed that selling access to Llama was not Meta’s business model. The court filing was part of a copyright lawsuit against Meta, which alleges that the company trained its Llama models using pirated ebooks. The filing states that Meta shares a portion of the revenue generated by companies that host Llama models for users. However, the document does not specify which hosts are paying Meta. Meta has partnerships with various Llama host partners, such as AWS, Nvidia, and Google Cloud. Developers can run Llama models without using these hosts, but many provide supportive services that simplify the process. Zuckerberg has previously mentioned licensing access to Llama models and suggested that Meta might earn revenue through other methods, including business messaging and advertising. Zuckerberg emphasized that the improvements made to Llama models come from collaboration with the AI research community. He believes that working openly benefits Meta's products. The filing's revelations are significant, as plaintiffs in the lawsuit accuse Meta of using pirated works to develop Llama and sharing these works via torrenting. Looking ahead, Meta plans to increase its capital expenditures significantly this year, focusing on AI investments. The company is considering a subscription service for Meta AI to help offset the costs associated with these plans.