Google was forced to restrict Meta's access to its advanced Gemini artificial intelligence models. According to the Financial Times, the move was triggered by massive internal demand for computing resources, which exceeded the search giant's available capacity.

The restrictions, introduced back in March 2026, have led to noticeable delays in several of Meta's internal AI projects. Due to the capacity shortage and as part of a broader cost optimization program, Meta's management urged employees to use AI tokens — units measuring model consumption — more sparingly. According to sources, although the shortage affected other Google clients as well, Meta suffered the most due to its extremely high demand for Gemini's capabilities.

The situation clearly demonstrates that even the world's largest developers are facing an acute infrastructure deficit. Google CEO Sundar Pichai confirmed that the company is constrained by computing resources, noting that Google Cloud revenues could have been significantly higher if it could meet all incoming demand. To address this problem, Google signed a massive agreement with SpaceX this month to lease computing power worth over $30 billion. Anthropic also recently signed a similar contract with SpaceX, and discussions are underway about creating orbital data centers.

Previously, Meta actively used Gemini for automating security systems, content moderation, support chatbots, and advertising services. However, the company is now gradually reducing its reliance on third-party solutions, transitioning to its own Muse Spark model, developed by Meta's Superintelligence Labs division. According to insiders, the new model already competes effectively with Gemini. In parallel, Meta is investing heavily in building its own infrastructure, planning to invest up to $600 billion in U.S. data centers by 2028.

Against this backdrop, Google continues to strengthen its AI presence, integrating Gemini models into classified Pentagon systems and unveiling its latest developments at the Google I/O 2026 conference, including Gemini 3.5 and the AI agent Spark.