Google has limited Meta's access to Gemini models due to a shortage of computing capacity

Google has reportedly limited the availability of its Gemini artificial intelligence models for Meta, not for commercial reasons but due to inadequate computing capacity to meet the demands of the group led by Mark Zuckerberg. This information comes from the Financial Times and highlights how the demand for dedicated AI infrastructure is outpacing supply even among major tech companies.

According to the report, in March, Google informed Meta that it could not provide all the processing capacity required to use Gemini. The reduction in available resources has also impacted some of the company's internal projects, causing delays in development. The scarcity of capacity has also prompted Meta to encourage its employees to use the so-called AI tokens more efficiently, which are the resources expended during the use of AI models. According to the published information, Meta was not the only company affected by these limitations. Other clients reportedly received fewer resources than their initial requests, but Meta's case was particularly significant due to its extremely high demand for computational capacity.

Meta, Gemini, and Infrastructure Challenges

The situation confirms how the availability of infrastructure and computing power represents one of the main obstacles to the development of artificial intelligence. Alphabet had already announced in April a 63% year-on-year growth in Google Cloud revenues in the first quarter, reaching $20 billion, largely due to the demand for AI-dedicated infrastructure and services.

To increase its processing capacity, Meta had already signed a multibillion-dollar agreement with AWS and a significant deal with AMD for the supply of GPUs for AI-dedicated data centers. Despite the current difficulties, both Google and Meta are significantly increasing their investments in infrastructure. Last November, Meta announced a plan exceeding $600 billion in the United States by 2028 to support research, infrastructure, and employment in the AI sector.

Additionally, in the first quarter of the year, it revised its capital expenditure forecasts for 2026 upwards, bringing them to between $125 and $145 billion, citing rising component prices and higher costs for new data centers. The situation also highlights a significant aspect: despite Meta long promoting the Llama family of models as an open alternative to other proprietary solutions, the use of Gemini models remains sufficiently important to influence some internal activities.