Google Has Imposed AI Limits on Meta Due to a Shortage of Computing Resources — FT

Google has imposed limits on social media giant Meta’s use of its Gemini AI models, the Financial Times reports, citing sources. According to the newspaper, Mark Zuckerberg’s company requested more computing power than Google was able to provide. This is yet another sign of the infrastructure shortage facing even the world’s largest AI developers, the newspaper notes.

Sources told the FT that Google notified Meta around March that it would be unable to provide the full amount of Gemini computing power that Meta had intended to purchase. This led to disruptions and delays in the implementation of some of Meta’s internal AI projects, the publication reports.

Due to current restrictions, as well as a general push to optimize operational expenses, Meta’s management has urged employees to use AI tokens—units that measure the volume of neural network usage—more efficiently, several sources told the FT.

In addition, several of Google’s other clients have also been affected by the restrictions, albeit to a lesser extent, according to the Financial Times, though it does not specify which companies are involved. Meta was hit harder than the others because of its exceptionally high demand for the search giant’s models, the FT notes.

Google and Meta declined to comment to the publication.

Shares of Alphabet, the holding company that owns Google, rose 1.1% in premarket trading on June 29, while Meta shares rose 1.5%.

Google’s decision to restrict a major client’s access to its models offers an inside look at the infrastructure pressures and capacity shortages mounting in the artificial intelligence industry, the FT notes. Demand for AI computing power has surged as companies integrate chatbots, code-writing assistants, and AI agents into their business processes. The resulting increase in inference workloads—generating responses to queries after AI models have finished training—has become one of the industry’s main challenges, the newspaper explains.

Despite spending tens of billions of dollars on chips, data centers, and electricity, even the largest tech companies are struggling to provide the computing power needed to meet the rapidly growing demand for cutting-edge models and AI services, the newspaper reports.

Due to such high demand, particularly from large corporate clients like Meta, Google was forced to urgently seek out additional resources. In early June, the company signed a $920 million-a-month agreement to lease computing power from SpaceX, a company owned by Elon Musk.

Unlike Google, Meta does not have its own cloud business, but it is rapidly expanding its data center infrastructure to meet its own needs for model training and inference. As part of this initiative, Meta has committed to investing $600 billion in the U.S. by 2028.

Gemini is used internally at Meta as part of a program to automate certain security processes—such as fraud detection and the removal of malicious content—as well as to power customer support and advertising chatbots. The model is also used for internal workflows and code writing alongside other solutions, including Claude from Anthropic, according to the FT.

This article was AI-translated and verified by a human editor