Google was unable to provide the Gemini computing power that Meta requested, causing Meta's internal project to be interrupted and delayed.
It has been reported that Google has restricted Meta's access to Gemini. This has become a hot topic as an example showing that even one of the world's largest AI companies faces infrastructure limitations.
Google caps Meta's Gemini use as AI demand strains capacity
According to sources familiar with the situation, there was originally an agreement between Meta and Google regarding the use of Gemini, but Google was unable to provide the computing power that Meta desired, resulting in restrictions on its use. As a result, some of Meta's AI projects, which were proceeding with the assumption that they would fully utilize Gemini's computing power, have been delayed.
This disruption, which began around March 2026, is still ongoing as of the end of June. Coupled with Meta's efforts to reduce its AI costs, Meta is reportedly urging employees to use AI tokens more efficiently.
A source stated, 'Other Google customers are also affected by the restrictions, but not to the same extent as Meta. Meta is particularly hard hit because there is a very high demand for Google's models.'
In its April 2026 earnings announcement, Google revealed that its cloud business revenue exceeded $20 billion (approximately 3.2 trillion yen) for the first time, and that its backlog of cloud contracts that have been signed but have not yet been delivered has nearly doubled compared to the previous quarter, reaching over $460 billion (approximately 74.4 trillion yen).
However, they are struggling to secure infrastructure, and demand from large corporate clients like Meta is directly straining Google's computing power. In June 2026, it was revealed that Google had secured computing power by contracting with SpaceX.
Google signs agreement to pay nearly 150 billion yen per month to access SpaceX's computing power - GIGAZINE
Meta is collaborating with Google, as well as semiconductor companies such as AMD and NVIDIA, to realize its AI strategy called 'Personal Superintelligence.'
Meta has agreed to purchase up to 6GW of AMD Instinct GPUs for over $100 billion, with 1GW to be deployed first in 2026 - GIGAZINE
The Financial Times wrote, 'Google's decision to limit model usage by large clients is a rare example that offers a glimpse into the growing infrastructure strain and bottlenecks across the AI industry. Even the largest technology companies are struggling to secure the computing power to support advanced models and AI services, despite investing heavily in chips, data centers, and power.'
Related Posts:
in AI, Posted by log1p_kr