Meta
has already invested millions of dollars into creating and developing open-source AI models in the Llama family, but training large LLMs is becoming increasingly expensive. So, as The Information reports, Meta has asked its “colleagues” Microsoft and Amazon for help, which is odd, since Meta has vast computing resources of its own.
Four sources said Meta offered these companies some kind of “incentives,” such as the ability to influence the development of future Llama AI models. It’s not entirely clear what Microsoft and Amazon benefit from this, since they are working on AI projects themselves, mostly commercial ones, while Meta makes money primarily from advertising. In addition, Llama is an open-source project, so sponsoring the creation of a strong competitor in the market is not very profitable.
In addition to the “most serious” discussions with Amazon and Microsoft, there have also been talks with Databricks, IBM, Oracle, and representatives of at least one Middle Eastern investor, according to at least two sources. Meta has also been in discussions with companies about forming a Llama Consortium group in early 2025. In those discussions, it is rumored to have primarily asked for money, as well as servers or other resources to offset the cost of training the models.
Image source: Juan Camilo Guarin P/unsplash.com
Other forms of support for potential partners were also discussed. For example, high-ranking Meta representatives could participate in partner conferences or provide more details on the model training process. The final results of the negotiations have not been published, but in the foreseeable future, it will probably be possible to track which competitors’ events Meta representatives will actively participate in.
In November 2024, it was reported that more than 100,000 NVIDIA H100 accelerators were used to train the then-new Llama-4 model. And in March, NVIDIA announced Llama Nemotron AI models with adjustable “intelligence level” based on Meta solutions.