Official announcement from Meta or reputable news sources

Meta Nears Completion of AI Cluster with 100,000 Nvidia H100 GPUs for Llama 4

Sep 11, 2024, 04:00 PM

Meta is close to completing an AI supercomputing cluster consisting of over 100,000 Nvidia H100 GPUs. This cluster will be used to train the Llama 4 model. Unlike previous setups, Meta will not use proprietary Nvidia networking gear for this project. This development places Meta among the few companies with such large-scale AI infrastructure, similar to Elon Musk's ventures. The cluster is also part of The Information's AI Data Center Database.