Google Capped Meta’s Gemini Access After It Requested More Compute Than Available

Meta wanted more Gemini computing power than Google could provide around March, and Google said no, capping the social media giant's access and delaying internal AI projects, the Financial Times reported Sunday. The restrictions hit Meta harder than other Google Cloud customers because of the sheer volume of compute it requested. Several other Google clients also faced capacity limits, but their smaller demands meant less disruption.

The shortfall forced Meta to tell employees to use AI resources more efficiently, including cutting back on "token" consumption, the units that measure how much computing power AI models consume. Staff were instructed to reduce unnecessary usage, according to the report.

Neither Google nor Meta commented on the story. The episode highlights a paradox playing out across the industry: even as tech companies pour billions into chips and data centers, demand for AI compute is outrunning supply. Google Cloud generated $20 billion in revenue in the first quarter ended March, but CEO Sundar Pichai said capacity constraints prevented even stronger results and caused the cloud unit's backlog to nearly double quarter over quarter.

Meta continues to invest heavily in AI across Facebook, Instagram, WhatsApp, and its expanding suite of AI tools. But the company is competing for infrastructure against the same hyperscalers (Google, Microsoft, and Amazon) that also supply it with cloud compute. For Google, the situation cuts both ways.

Gemini is a product it sells to enterprise customers through Google Cloud, and turning away a buyer as deep-pocketed as Meta means leaving revenue on the table. But Google's own infrastructure is strained, and Pichai has acknowledged that compute shortages are limiting the cloud business from growing even faster.