Google Throttles Meta's Access to Gemini as AI Infrastructure Demand Outstrips Supply

The Financial Times reports that Google has begun capping how much Meta can access its Gemini models — not over contract disputes, but because there is simply not enough compute to go around. It is the first documented case of one hyperscaler throttling another over AI infrastructure limits, and it signals that the demand side of the AI boom has officially outrun the supply side.

For Meta, the squeeze is both embarrassing and clarifying. The company has bet heavily on Gemini for parts of its AI product stack, but rationed access could accelerate its pivot toward in-house models — Llama in particular. CEO Mark Zuckerberg has long positioned open-source LLMs as the company's strategic hedge; Google's capacity cap just made that hedge look prescient.

The broader implication is harder to ignore. If two of the world's best-funded technology companies cannot reliably supply each other with inference capacity, the AI infrastructure shortage is real — not just a talking point for chipmakers selling GPUs. Every enterprise building on third-party AI APIs should be watching this closely.

There is something darkly funny about the company famous for infinite scale having to tell another tech giant to slow down.

Read the full story at Financial Times