Surging API requests have forced Google to implement compute-based usage limits, with implications for how tech giants share AI infrastructure
Google is tightening the spigot on its Gemini AI platform as demand from developers, enterprises, and fellow tech giants threatens to overwhelm available capacity. The move comes as Gemini API requests more than doubled between March and August 2025, forcing Google to rethink how it allocates one of the most sought-after resources in tech: raw AI compute.
Among those feeling the squeeze is Meta, which has been in discussions with Google Cloud about leveraging Gemini models for its advertising business. The situation illustrates a strange new dynamic in Silicon Valley, where bitter rivals are quietly becoming each other’s customers in the AI arms race.
What Google actually changed #
Starting May 17, 2026, Google imposed compute-based usage limits on Gemini Apps. Think of it like a cellular data plan: instead of unlimited requests, users now operate under rolling 5-hour refresh windows and weekly caps.
The limits apply broadly, not just to one company. Google has documented rate limits and spending tiers designed to ensure fair API usage across all customers during what the company characterizes as a rapid growth phase.
The company has reported continued sales growth for its Gemini AI products into 2026, suggesting that the demand constraints are a consequence of commercial success rather than any technical failure.
The Meta angle #
The relationship between Google and Meta here is fascinating, bordering on surreal. These are companies that compete fiercely for digital advertising dollars. Yet in September 2025, Meta was actively discussing how to integrate Google’s Gemini AI to improve ad targeting based on Meta’s own user data.
The compute-based limits Google introduced apply to all users through documented rate limits and spending tiers, creating a system where access scales with how much you’re willing to pay and how much capacity is available. There’s no public evidence that Google designed bespoke restrictions targeting Meta specifically. The constraints appear to be a blunt instrument applied across the board.
Why this matters beyond two tech giants #
The doubling of Gemini API requests between March and August 2025 suggests adoption curves for frontier AI models are steeper than many forecasts anticipated.
The fact that Google is implementing usage caps rather than simply scaling up infrastructure tells you something about the economics. Even for a company with Google’s resources, the capital expenditure required to meet unconstrained AI demand is daunting enough to warrant demand management in the interim.
The competitive dynamics get even more interesting when you consider that Meta has its own open-source Llama models. The fact that Meta is exploring Gemini integration despite having in-house alternatives suggests that Google’s models offer something Meta’s don’t, at least for certain use cases.
Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our