Google limits Meta’s Gemini AI access amid compute crunch

This story crystallises one of the defining constraints in the race to AGI: usable compute, not just clever model architectures. Google telling Meta it cannot supply the Gemini capacity requested, and nudging the company to ration AI tokens internally, shows that even the largest hyperscalers are hitting real infrastructure ceilings. For all the talk of infinite scalability, GPU clusters, power, and cooling are bottlenecks that force firms to prioritise which workloads get the best models. ([indianexpress.com](https://indianexpress.com/article/technology/artificial-intelligence/google-caps-metas-use-of-gemini-ai-models-report-10762613/))

Strategically, this concentrates power in whichever players can lock in long-term access to compute—whether through in‑house data centres, exclusive cloud deals, or alternative providers like CoreWeave. Meta’s dependence on Google’s models, Apple’s planned use of Gemini for Siri, and Microsoft’s own guidance to staff about third‑party tools like Anthropic’s Claude all show that the AI stack is blurring between rivals and suppliers. In the short term, these capacity caps may slow individual projects; in the medium term, they will likely trigger another wave of capex, long‑dated offtake contracts, and diversification toward open-weight models that are cheaper to run.

For the AGI race, the signal is that simply “throwing more tokens at the problem” is no longer free. Efficiency, architecture choices, and multi‑cloud optionality will matter more, while smaller players without privileged compute access may be pushed further to the margins.