Thought Leadership
Google has restricted Meta’s access to its Gemini AI models. The Facebook parent company had been using the external AI technology internally without public disclosure.
Google has limited Meta’s use of its artificial intelligence platform Gemini. According to reports by the Financial Times, Google informed its partner company around March 2026 that it could not provide the requested capacity in full due to extremely high demand for computing resources. The restriction remains in place and has caused delays and disruptions for Meta’s internal AI projects. As a result, Meta management instructed employees to use so-called tokens more efficiently. These tokens represent the data units processed by AI models.
Hidden use across Meta’s core business areas
Although Meta is investing billions of dollars in developing its own Llama model family, the company relies on competing AI technology from Google for important internal operations. According to the reports, Meta used Gemini models for customer service, chatbots for advertisers, software development tasks, as well as fraud detection and the removal of harmful content.
The decision was driven by Gemini’s stronger performance compared with Meta’s own systems. In addition to Google’s technology, Meta also appears to be using models from AI provider Anthropic. Unlike Google, Microsoft, or Amazon, Meta does not operate its own cloud infrastructure and therefore depends on external computing capacity.
Global data centers operating at full capacity
The shortage is not limited to Meta. Other Google Cloud customers are also affected, although to a lesser extent. Demand for computing power is currently growing faster than new data centers can be built.
In the last quarter, Google’s own AI models processed more than 16 billion tokens per minute through the company’s direct API access, representing a 60 percent increase compared with the previous quarter.
Google Cloud’s order backlog has grown to more than $460 billion US dollars. In the long term, Meta aims to reduce its dependence on competitors by expanding its own data center infrastructure and developing customized MTIA accelerator chips in cooperation with Broadcom.
(ll)
