Google is limiting access to Gemini for Meta due to insufficient computing resources.

      In brief, Google has restricted Meta’s access to the Gemini AI models due to limitations in computing capacity. In response, Meta has instructed employees to utilize AI tokens more efficiently and is transitioning to its own Muse Spark model.

      According to a report by the Financial Times, Google has imposed limits on Meta's utilization of its Gemini AI models as it is unable to meet the computing power requirements that Meta desired. This restriction has impacted various Google clients, with Meta being particularly affected.

      The situation has also influenced Meta's internal initiatives. As reported by three sources familiar with the situation, employees have been advised to optimize their use of AI tokens. Both Google and Meta have refrained from commenting on the news.

      Meta initially depended on Gemini, which outperformed its own Llama open-source models, for automating safety measures such as removing harmful content and combating scams. However, the company has been progressively transferring workloads to Muse Spark, a new internal model, in an effort to reduce reliance on external AI sources. Google itself is facing significant computing constraints and has agreed to pay SpaceX $920 million monthly for access to 110,000 Nvidia GPUs, referring to it as “bridge capacity” to meet escalating demand for Gemini Enterprise.

      This development highlights how the AI compute shortage is redefining interactions among major companies in the industry. Despite owning one of the largest AI infrastructure pools globally and allocating over $180 billion for capital expenditure this year, Google still cannot satisfy all customer demands. The rationing of access to a significant company like Meta, while concurrently leasing GPUs from a space company, underscores that infrastructure expansions are not keeping pace with consumption.

      For Meta, relying on a rival's AI models has always been an uneasy situation. The company eliminated 8,000 jobs in May and shifted billions toward AI infrastructure, with capital expenditure projections between $115 and $135 billion for 2026. It has reassigned 7,000 employees to AI-oriented positions and initiated Muse Spark through its Superintelligence Labs division. The restrictions on Gemini expedite a transition Meta was already undertaking, moving away from dependence on external leading-edge models to developing internal solutions that can manage critical operations like large-scale content moderation.

      The broader trend is evident across the entire industry. The demand for AI compute resources is escalating more rapidly than even the most ambitious infrastructure investments can accommodate. Google is acquiring capacity from SpaceX, Anthropic is leasing an entire data center from SpaceX, and Meta is being advised to minimize its token usage by its own cloud provider. The most pressing bottleneck in the AI boom lies not in algorithms or talent but rather in the physical infrastructure needed to support them.