Google Informs Meta of Limits on Gemini Access Due to Insufficient AI Capacity
Google has apparently restricted Meta’s access to its Gemini AI models after the Facebook parent company requested more computing capacity than the search giant could provide. According to a famous publication report, Google informed Meta around March that it could not meet the full demand for Gemini AI capacity due to infrastructure limitations, a move that delayed some of Meta’s internal AI projects.
These restrictions—which reportedly remain in effect—come at a time when AI companies are facing a growing infrastructure challenge. Even companies investing tens of billions of dollars in chips, data centers, and energy are struggling to secure the computing capacity needed to meet the rising demand for AI models and services.
The report, citing people familiar with the matter, notes that Meta has been one of Google’s largest Gemini customers and has been more affected than most due to its exceptionally high demand for AI computing resources. While a small number of other corporate clients have faced similar restrictions, the sheer volume of Meta’s usage has made it particularly vulnerable to this capacity shortage.
The report indicates that Google told Meta it simply did not have enough computing resources to provide the full Gemini capacity the company requested. This shortage has reportedly disrupted and delayed several of Meta’s internal AI initiatives.
Meta Restricts Employee AI Usage
Consequently, Meta has also begun restricting how its employees use AI internally. Reportedly, the company has encouraged its staff to be more efficient with AI tokens—the units used to measure AI model usage—as part of a broader effort to cut AI costs while operating under new capacity limits.
Why does Meta use Gemini instead of Llama?
Gemini has apparently become a key tool within Meta. The company uses Google’s AI models to automate security operations, including scam detection and the removal of harmful content. Gemini also powers certain internal customer support tools, advertising assistants, coding workflows, and productivity tasks, alongside models like Anthropic’s Claude. According to the report, Meta initially adopted Gemini because it outperformed the company’s own open-source Llama models in various business use cases. However, the company has recently begun shifting some workloads to its new Muse Spark model; internal sources believe it is already competitive enough to reduce Meta’s reliance on third-party AI providers.
Meanwhile, Google is also facing a capacity shortage.
These constraints highlight the growing infrastructure bottleneck affecting the entire AI industry. While companies have focused primarily on developing increasingly capable AI models, running those models at scale has become a challenge of equal magnitude. Demand for inference computing—the processing power required whenever an AI model responds to a query or performs a task—has surged as companies deploy chatbots, coding assistants, and AI agents within their operations.
To meet this demand, Google has moved quickly to expand its infrastructure. The company reportedly signed a deal worth approximately $920 million per month to lease additional computing capacity from Elon Musk’s SpaceX. Similarly, the AI startup Anthropic has reportedly secured a comparable infrastructure agreement.
Google itself has previously acknowledged that computing capacity remains a limitation. During the company’s first-quarter earnings presentation in April, CEO Sundar Pichai noted that Google Cloud revenue surpassed $20 billion for the first time, though he added that the figure could have been even higher had Google possessed sufficient infrastructure to meet customer demand.