The bottom line: Meta is dependent on AI capacity from Google’s Gemini despite the Facebook parent company developing its own language models, and is suffering from throttling due to global computing resource bottlenecks.
Google has throttled Meta’s access to its Gemini models since March 2026 due to high global demand for computing power. Meta used the Google systems internally for critical functions such as customer service and fraud detection — despite the company simultaneously investing billions in its own Llama models.
Google informed Meta around March 2026 that the provisioned capacity for Gemini would be reduced due to extreme demand for computing power. This throttling persists and leads to delays and disruptions in Meta’s internal AI projects. Management instructed staff to use tokens — processed data units — more sparingly.
Meta used Google’s Gemini models for customer service, advertising customer chatbots, programming tasks, fraud detection, and filtering harmful content. The higher performance of Gemini compared to Meta’s own Llama models was the reason for this use. In addition to Google, Meta also accesses models from Anthropic. Because Meta, unlike Google, Microsoft, or Amazon, does not operate its own cloud infrastructure, the company is dependent on external capacity.
The bottleneck also affects other Google Cloud customers, albeit to a lesser extent. In the last quarter, Google’s models processed more than 16 billion tokens per minute via the API — an increase of 60 percent compared to the previous quarter. Google Cloud’s backlog grew to over 460 billion US dollars. Demand for computing power is growing faster than new data centers can be built.
Meta is attempting in the long term to reduce this dependency by expanding its own data centers and developing specialized MTIA accelerator chips in collaboration with Broadcom.
Source: www.it-daily.net · Published June 30, 2026
Lumi AI News — AI-assisted curation pursuant to Art. 50 EU AI Act. Paraphrase and classification by Lumi News Pipeline v1.7.2.