Due to the ongoing global shortage of cloud computing capacity, Google has officially restricted Meta's access to its top AI model, Gemini. As a core component of Meta's automated security workflows, Gemini was widely used for large-scale review tasks such as fraud detection and harmful content filtering, and its efficiency had once surpassed Meta's own open-source Llama system.

gemini, Google

However, with the surge in AI inference workloads, even though Google achieved $20 billion in cloud business revenue in the first quarter, the speed of its physical infrastructure construction still failed to keep up with the explosive growth of computing demand. This allocation of computing power by Google to Meta has caused delays in several internal AI projects at Meta.

Faced with this sudden infrastructure bottleneck, Meta's management has urgently asked employees to improve the efficiency of AI tokens. At the same time, this situation constrained by a competitor has accelerated Meta's move towards independence. Under the push of its newly established "Super Intelligence Lab," Meta is migrating its core security and review workloads to its fully self-developed cutting-edge model, "Muse Spark."

This incident of restricted computing power reveals the core contradiction in the deepening development of the AI industry: the bottleneck limiting AI advancement is no longer algorithms or talent, but the scarcity of physical resources such as chips, electricity, and data centers. It is forcing tech giants to accelerate their shift from relying on external cloud ecosystems toward costly infrastructure self-building.