Google Throttles Meta’s Gemini Access

Alphabet’s cloud arm can’t satisfy Meta’s hunger for AI compute and is curbing its access. The twist: Meta already runs its own frontier model, Muse Spark — while rival Apple pays Google billions for Gemini.

Google has restricted Meta’s access to its Gemini AI models. That’s according to the Financial Times, citing people familiar with the matter. The trigger, the report says, was that Meta requested more computing capacity than its rival could supply.

Around March, Google — a subsidiary of Alphabet — reportedly told Meta it could not fully provide the Gemini capacity the company had sought to buy. The shortfall disrupted and delayed some of Meta’s internal AI projects. Other Google clients have been affected too, though to a lesser extent; Meta has been hit particularly hard, the report says, because of its exceptionally high demand for Google’s models. In response, Meta has reportedly urged staff to be more sparing with AI tokens — the units that measure AI usage.

Neither Google nor Meta has officially confirmed the account; both companies initially did not respond to requests for comment. The news agency Reuters said it could not independently verify the report either.

The Crunch Is Industry-Wide

The case illustrates a structural problem across the entire sector: despite billions invested in chips and data centers, the major tech companies are still struggling to secure enough computing power to meet the growing demand for AI services. At Google Cloud, revenue climbed to 20 billion US dollars in the first quarter (ended March). CEO Sundar Pichai noted, however, that capacity constraints prevented even stronger growth — the cloud unit’s backlog had nearly doubled quarter on quarter.

A Pointed Backdrop: Meta Has Its Own Model

What makes Meta’s demand notable is that the company is by no means without AI technology of its own. In April 2026, the newly formed Meta Superintelligence Labs under Alexandr Wang unveiled Muse Spark — Meta’s first proprietary frontier model and a break with the open-weights strategy of its Llama line. Muse Spark is built as a multimodal system with reasoning and multi-agent orchestration, and it now powers Meta’s AI assistant in the Meta AI app as well as on WhatsApp, Instagram, Facebook, and Messenger. In independent benchmarks, the model ranked among the leaders, behind Google’s Gemini, OpenAI’s GPT, and Anthropic’s Claude.

That Meta nonetheless sought to buy Gemini capacity from Google on a large scale shows how far the need for AI inference can outstrip a company’s own model development — and how intertwined the supply chains of competing tech giants have become.

Apple Pays, Meta Gets Throttled

A second prominent Gemini customer sharpens the contrast: Apple. In January 2026, the iPhone maker struck a multi-year partnership with Google under which a custom version of Gemini powers its overhauled “Siri AI.” According to Bloomberg, Apple pays roughly one billion US dollars a year for it. The model weights run on Apple’s own Private Cloud Compute infrastructure; by Apple’s account, Google receives no user data.

That means money flows in both directions: Google has long paid Apple an estimated 20 billion US dollars a year to be the default search engine on iPhones — and now Apple pays in return for Google’s AI. While Apple’s billion-dollar contracts are being honored, Meta is running into capacity limits with the very same provider. Whether this is purely a matter of available compute or also of strategic priorities within the customer portfolio, the companies involved left open.