Meta Platforms has reportedly agreed to a multibillion-dollar deal to rent Google Cloud's custom artificial intelligence chips. These Tensor Processing Units (TPUs) will be used by Meta to train and run its next-generation large language models.
This move signifies Meta's strategy to diversify its AI hardware supply chain, reducing reliance on Nvidia, which currently dominates the market. Google is leveraging its advanced TPUs, including the recent Ironwood models, as a compelling, lower-cost alternative to Nvidia's GPUs.
Meta is actively seeking to avoid dependence on a single chip provider. The company has recently announced substantial deals with both AMD for AI chips and now Google for cloud-based TPUs. This multi-supplier approach allows Meta to pair optimal silicon with specific AI workloads and potentially secure better pricing.
Google aims to challenge Nvidia's market dominance by offering its TPUs not only through cloud rentals but also for direct sale to customers for private data centers. This expansion could see Google capture a significant portion of Nvidia's data center revenue.