Google capped Meta’s use of Gemini

Google limited Meta’s access to Gemini because Meta wanted more computing capacity than Google could supply. The limit was communicated around March 2026 and disrupted or delayed some of Meta’s internal AI projects. Meta told staff to use tokens more carefully while also trying to control AI costs.

Meta has used Gemini and Claude for internal work such as scam detection, harmful content handling, customer support, ad helper chatbots, workflow support, and coding. Gemini became important inside Meta because it performed better than Meta’s own Llama models for some tasks. Google has also been short on capacity and signed a deal to rent computing power from SpaceX for about $920 million per month.

Google Cloud passed $20 billion in quarterly revenue, but its backlog of signed work grew to more than $460 billion, and Sundar Pichai said limited computing capacity held back revenue. In the same competitive setting, Meta also ran a large safety benchmarking project against ChatGPT, Gemini, and Character.AI using teen-like accounts and prompts about suicide, sex, drugs, and eating disorders; rival companies said the testing was not authorized or may have broken their rules.