Google Reportedly Caps Meta's Gemini AI Usage Amid Compute Shortage

The ongoing "AI arms race" has hit a significant bottleneck: computing power. A recent report from Engadget reveals that Google has reportedly capped Meta's access to its powerful Gemini AI models, a decision driven by the immense and growing demand for AI infrastructure. This restriction has forced Meta to re-evaluate its internal AI operations and has reportedly caused delays in various projects. Meta, despite developing its own suite of Llama open-source models, has been a substantial user of Google's Gemini AI. The social media giant leveraged Gemini for a wide array of critical internal functions, including sophisticated fraud detection, content moderation, powering customer service chatbots, and providing coding assistance to its developers. Insiders familiar with the situation indicated that Meta chose Gemini for these tasks because it consistently outperformed its proprietary Llama models. The capacity crunch became apparent when Google, in March, informed Meta that it could not fulfill the full computing capacity requested. This warning led to Meta implementing internal directives for its engineers to use AI tokens more efficiently, effectively rationing their access to Google's AI services. This situation underscores a broader industry challenge where even companies with vast resources, like Meta, are struggling to secure adequate computational power for their AI ambitions. In response to this growing reliance on external providers and the scarcity of compute, Meta has committed to massive infrastructure investments. Mark Zuckerberg has pledged approximately $600 billion through 2028 to construct new data centers across the United States. Unlike Google or Microsoft, Meta does not operate a commercial cloud business, meaning these substantial hardware investments are solely for its internal services and to reduce its dependence on competitors. The article also points to the broader industry-wide struggle for compute. Google itself, facing unprecedented demand for its Gemini Enterprise services, has reportedly entered into a staggering $920 million-a-month agreement with Elon Musk's SpaceX to lease additional data center capacity from xAI. Similarly, rival AI lab Anthropic is said to have secured a comparable emergency capacity deal with SpaceX last month. This fierce competition for GPUs and data center space illustrates that the bottleneck is not just at the model layer but extends deeply into the foundational infrastructure required to run advanced AI.