Meta faced a significant capacity crunch around March, unable to meet the full Gemini capacity it had sought to purchase. This shortfall disrupted and delayed some of Meta’s internal AI projects, as reported by the Financial Times.
The issue highlights a growing infrastructure bottleneck in the AI industry, where demand for computing power now outpaces available resources, affecting even the largest technology companies.
Meta is particularly affected due to its high demand for Google models, leading it to adopt efficiency strategies within its own system. Employees had to be more conservative about using “tokens,” which measure how much text an AI model processes, and several internal projects were reportedly delayed.
While other Google Cloud customers also experienced similar shortages of capacity, none faced such enormous demands for artificial intelligence resources as Meta did.
Google Cloud’s revenue rose to $20 billion in the first quarter ending March. CEO Sundar Pichai noted that computing power shortages had held back revenue growth and contributed to a nearly doubling of the cloud division’s backlog in just one quarter.
Despite significant capital expenditure on specialist chips and data centers, firms have been unable to build fast enough to keep up with rising demand for AI products. Starting May 17, 2026, Google introduced computing-based limitations for Gemini Apps, which saw an increase in API requests from March to August 2025.