Google Caps Meta's Access to Gemini AI Amid a Compute Crunch

Google told Meta this spring that it could not supply all the computing capacity the social-media giant wanted to run Google's Gemini AI models, according to reporting by the Financial Times — a squeeze that disrupted some of Meta's internal projects and exposed the limits of even the world's largest AI infrastructure.

Anjali Rao June 29, 2026 at 7:48 AM

In the AI arms race, the surprise is not who is buying — it is who has run short of supply.

Google told Meta no

Around March, Google delivered an awkward message to one of its biggest cloud customers: it could not provide all the Gemini AI capacity that Meta had sought to buy. The Financial Times first reported the squeeze on June 28, citing people familiar with the matter, and CNBC and Reuters confirmed the broad outlines. Neither Google nor Meta commented when contacted by reporters.

The shortfall disrupted and delayed several of Meta's internal AI initiatives, according to the FT's sources, prompting Meta to tell staff to use their AI "tokens" — the units that measure how much computing an AI task consumes — more sparingly. In effect, one of the richest companies in technology put its own employees on a compute diet.

A peculiar dependency

The episode carries some irony. Meta is among the industry's most aggressive AI self-builders: it develops the open-source Llama models and has committed to spending on the scale of tens of billions of dollars a year on AI infrastructure. Yet it was still leaning heavily on Google's Gemini for internal work, at a volume even Google's cloud division could not fully meet — a reminder that, at the frontier, no one is entirely self-sufficient.

Google's own capacity problem

By Google's telling, the restriction is not competitive strategy but a supply constraint. Google Cloud reported $20 billion in revenue for the quarter ended March, a figure Chief Executive Sundar Pichai said could have been higher but for computing limits, and the unit's order backlog grew sharply. Other Google Cloud customers faced similar but milder limits, the FT's sources said; Meta stood out for the sheer scale of its demand. (Some figures circulating about Google's stopgap measures and Meta's internal model plans come from single, unconfirmed reports and are not established.)

What it signals for the AI race

The crunch matters beyond the two companies. It suggests the AI infrastructure build-out — with Google, Microsoft, Amazon and Meta all spending at historic rates — has not yet closed the gap between what customers want and what providers can deliver. Companies are signing large cloud commitments only to be told, months later, that the capacity is not there.

For Meta, the practical effect is more pressure to build in-house and lean less on outside providers for critical work. For Google, it is a rare moment in which a marquee customer's needs outran its ability to supply — a vulnerability its rivals will have noticed. As of publication, neither company had issued a public statement.