Report: Google Put Limits on Meta’s Heavy Gemini Usage After Token Surge

Meta’s aggressive use of Google’s Gemini AI reportedly hit a wall earlier this year, after Google moved to cap how much the social giant could tap its models.

According to a report in the Financial Times, cited by Gizmodo, Google told Meta in March that it could no longer keep pace with the company’s demand for Gemini and imposed limits on its access. The story is based on unnamed sources “familiar with the matter.”

In March, Meta emerged as one of the most enthusiastic participants in what was dubbed a “tokenmaxxing” trend, a workplace push where employees were effectively judged by how many AI tokens they burned through on the job. Tokens are the basic units that measure AI usage, so more tokens typically mean more or larger model calls.

This moment overlapped with a separate fad inside the industry: so-called “token-hungry” agentic AI platforms such as OpenClaw. These systems were reportedly being used by software engineers hoping to boost productivity by offloading more of their workflow to AI agents.

The combination appears to have put serious strain on Google’s infrastructure. The Financial Times, citing three people familiar with the situation, reports that Google informed Meta it could not keep up with the level of AI usage and therefore introduced caps on how Meta could use Gemini.

Those restrictions reportedly disrupted and delayed some of Meta’s internal AI projects. With the new limits in place, and amid a broader push to control AI spending, Meta is said to have encouraged employees to use tokens more sparingly instead of chasing higher and higher token counts.

Meta was not alone in putting pressure on Google’s AI infrastructure, according to the same reporting. Other large customers are also said to have faced usage caps as Google tried to manage demand on Gemini. However, the Financial Times sources suggest Meta stood out, even among other heavy enterprise users.

The resource burden from this kind of large-scale consumption may have contributed to Google’s decision in early June to rent compute capacity from SpaceX, the parent company of xAI, in a deal reportedly priced at $920 million per month. The report does not directly tie that agreement to Meta’s usage, but presents the strain from big customers as part of the backdrop to Google’s infrastructure moves.

For now, the reported restrictions on Meta remain in place, and the era of internal “tokenmaxxing” there appears to have given way to more careful token accounting.