Google ограничила Meta доступ к Gemini из-за дефицита вычислительных мощностей

Google has limited Meta's access to its Gemini AI models — the company cannot keep up with demand for computational resources and is introducing quotas for corporate clients.

What

Happened On Sunday, Financial Times reported that Google sent usage restrictions for Gemini to a number of enterprise clients. The restrictions affected several major companies, but Meta suffered most severely: the social network was requesting a significantly larger volume of computational resources than Google was willing to allocate. A telling detail: Meta is a paying corporate client of Google — and even that did not help.

A situation where a major enterprise customer with a budget does not get the resources it needs clearly illustrates the scale of the shortage. It is not a market deficit, but a physical one: simply not enough servers, chips, and data center capacity. The situation is unfolding against the backdrop of record growth in AI inference demand — Google as an infrastructure provider is forced to distribute limited resources among thousands of clients.

How This Hit Meta The restrictions have already affected Meta's internal AI projects.

The company is running one of the most ambitious AI programs among technology corporations: artificial intelligence is built into Facebook, Instagram, WhatsApp, and Meta Ray-Ban smart glasses. In public statements, Meta has repeatedly emphasized its plans to embed AI in all its products — a disruption in computational resource delivery directly threatens this strategy. The blow is all the more painful because Meta is simultaneously developing its own family of open Llama models — and they too need capacity for training and inference: Delays in AI features could affect billions of users on Facebook, Instagram, and WhatsApp Training new versions of Llama requires massive computational volumes Meta's internal AI deadlines face the threat of missing Dependence on an external provider becomes a strategic vulnerability ## Why Google Lacks Capacity Over the past two years, demand for AI inference has grown radically.

Each new generation of models is significantly more powerful than the previous one — but also significantly more resource-hungry: modern flagship models consume many times more compute per request than their 2023 predecessors. Google is building new data centers and ramping up production of its own TPU chips, but infrastructure is not keeping pace with demand. Complicating the situation is that Gemini is now deeply embedded not only in corporate products for business clients, but also in Google's own internal services — Workspace, Search, Android.

The company's own needs compete with external customer requests for the same physical resources.

What

This Means Computational capacity shortage is becoming a new strategic constraint in the AI race. Companies without their own infrastructure risk ending up at the back of the line — even with a budget. For Meta, this is another signal to accelerate investments in its own data centers. For the market as a whole — a reminder that the competition for AI computing power in the coming years will be no less intense than competition for the models themselves. *Meta has been recognized as an extremist organization and banned in the Russian Federation.