Google Cuts Off Meta's Gemini Access as AI Compute Capacity Runs Short

Google has capped Meta's access to its Gemini AI models. The social media company wanted more compute capacity than Google could supply, according to a Financial Times report published Sunday . Alphabet told Meta around March it could not meet the full Gemini capacity Meta sought to purchase, disrupting and delaying some internal AI projects .

The restrictions hit Meta harder than other Google clients due to its exceptionally high demand for Gemini models. Several other customers have also been affected, though to a lesser extent . Neither Google nor Meta has commented publicly on the report.

Meta Asks Staff to Use AI Tokens More Efficiently

In response to the capacity limits, Meta has encouraged employees to be more efficient with AI tokens - the units that measure AI usage . The move comes as Meta pushes to reduce AI costs across the company while still pursuing aggressive AI product development.

The capacity crunch reflects a broader industry problem. Companies are spending billions on chips and data centres but still struggling to secure enough computing power to support surging AI demand .

Google Cloud's Own Growth Hampered by Compute Constraints

Google Cloud revenue grew to $20 billion in the first quarter ended March. But CEO Sundar Pichai said computing power constraints prevented even higher growth and contributed to the cloud unit's backlog nearly doubling quarter on quarter .

The shortage appears to be hitting AI inference hardest. OpenAI and Broadcom recently unveiled their custom "Jalapeño" inference chip, while Anthropic has been in talks to run Claude on Microsoft's Maia 200 silicon - both signs that major players are racing to secure compute outside traditional cloud channels.

A New Chapter in the Google-Meta Rivalry

The compute restrictions add a new dimension to the long-running rivalry between the two tech giants. In September 2025, Meta was reportedly in talks with Google Cloud to leverage Gemini for more effective advertising - exploring options to fine-tune Gemini and open-source Gemma models with its own ad data . Both companies compete directly in the online advertising market.

Meta has since pivoted toward building its own AI infrastructure stack. The company has spent years developing Llama, its open-source family of large language models, and designing custom MTIA accelerators to reduce dependence on Nvidia and external providers.

For most engineers, the real question is what happens when companies can no longer rely on their rivals for critical AI infrastructure. OpenAI vs Anthropic vs Google DeepMind already shows how the competitive landscape is fragmenting. Google's decision to throttle Meta suggests that as AI compute becomes scarcer, platform neutrality may be the first casualty.

What Happens Next

Google's capacity constraints show no signs of easing. The company's AI infrastructure chief told employees in November 2025 that Google must double its AI computing capacity every six months just to keep pace with demand . Meta's $145 billion capex forecast for 2026 reflects its bet on self-built infrastructure to avoid similar bottlenecks in the future . Neither company has indicated whether Gemini access will be restored or if the restrictions are permanent.