Google has imposed limits on Meta's utilization of its Gemini AI services, a move attributed to the surging demand for AI infrastructure and Google's strained capacity. This constraint has prompted Meta to seek alternative solutions for its artificial intelligence needs.
In response to these capacity limitations, Meta Platforms finalized a multi-billion dollar, multi-year agreement on February 28, 2026, to lease Google's Tensor Processing Units. This deal, estimated to be worth between $4 billion and $8 billion over the next three years, grants Meta priority access to Google Cloud's TPU v6 and v7 architectures. The primary goal for Meta is to train its upcoming Llama 5 and Llama 6 models, diversify its AI infrastructure, mitigate supply chain risks, and optimize training costs.
This partnership with Google does not replace Meta's existing relationships with Nvidia or its increasing use of AMD's Instinct accelerators. Instead, it establishes Google as a key infrastructure partner. By utilizing Google's custom-designed silicon, Meta aims to achieve better performance-per-watt and a lower total cost of ownership compared to general-purpose GPUs, addressing the ongoing challenge of meeting high AI compute demand.
