Google has restricted Meta’s access to its Gemini artificial intelligence models after demand from the tech giant outstripped available computing resources, Financial Times reports.
The restrictions, in place since March 2026, have delayed some of Meta’s internal AI projects and are yet another sign that infrastructure shortages have become one of the AI industry’s biggest challenges.
According to the publication, Google told Meta back in March that it would not be able to provide the full amount of compute needed to work with Gemini.
The restrictions remain in effect. Because of this, and as part of a broader cost-optimization program, Meta urged employees to use AI tokens more sparingly — a unit used to measure model resource consumption.
Sources told the publication that the restrictions also affected other Google customers, but Meta was hit the hardest due to exceptionally high demand for Gemini models.
The situation shows that even the world’s largest AI developers are running into infrastructure constraints despite multibillion-dollar investments in data centers, GPUs, and power capacity.
The rapid spread of chatbots, agents, content-generation services, and coding tools has significantly increased the load on inference systems — running models after their training is complete.
According to Google CEO Sundar Pichai, the company is already feeling a shortage of compute resources.
“Obviously, we are compute-constrained in the near term. And as an example, our Cloud revenue would have been higher if we were able to meet the demand,” he said during the quarterly earnings presentation.
To expand its infrastructure, Google this month signed a deal with SpaceX to lease computing capacity worth $920 million per month. The total will exceed $30 billion.
Anthropic — the developer of the Claude chatbot — recently signed a similar deal with SpaceX.
In addition, Google and SpaceX have begun talks on building orbital data centers for artificial intelligence.
Despite the rapid development of its own Llama ecosystem, Meta used Gemini in internal processes for a long time, as these models delivered better results.
In particular, Gemini was used for:
At the same time, the company is gradually switching to its own Muse Spark model, introduced after the reorganization of the Meta Superintelligence Labs unit. According to sources, the new model is already considered competitive with Gemini and makes it possible to reduce the company’s reliance on third-party AI providers.
Meta is also actively expanding its own infrastructure. Since the company does not have a cloud business like Google Cloud, it is investing significant funds in its own data centers. By 2028, Meta plans to invest $600 billion in the development of AI infrastructure in the United States.
Earlier, Google signed a deal with the U.S. Department of Defense to integrate Gemini models into the Pentagon’s classified systems. In addition, at the Google I/O 2026 conference, the company unveiled the Gemini 3.5 and Omni models, the Spark AI agent, and a number of other products, making artificial intelligence the central theme of its development.
For commercial:
Editorial:
Editor:
Ukraine, Mykolaiv, Lazurnaya str. 5/2
Official documents:
Leave a Reply