Google has imposed restrictions on Meta's use of its Gemini artificial intelligence models after the social media giant requested more computing capacity than Google was able to supply, according to a Financial Times report. The limits reportedly took effect around March 2025, marking a significant flashpoint in the intensifying global scramble for AI infrastructure.

What Triggered the Restrictions

Meta's demand for Gemini model capacity was reportedly so outsized that Google could not fulfil the full request, leaving a shortfall that disrupted and delayed several of Meta's internal AI projects. According to the report, Meta ranks among Google's largest cloud customers, making its exposure to any capacity ceiling particularly acute.

The constraints are said to have prompted Meta to direct employees to use AI resources more efficiently — including cutting consumption of AI tokens, the standard unit used to measure and bill generative AI usage.

Broader Impact Across the Industry

Meta is not alone. Several other Google customers have reportedly encountered similar limitations on computing access, though the disruption to them has been less severe. The pattern points to a structural bottleneck: across the technology industry, demand for AI computing power is consistently outrunning available supply, even as companies pour billions into data centres and next-generation chips.

This comes amid a broader infrastructure arms race, with hyperscalers racing to commission new GPU clusters and custom silicon. Notably, the constraints are emerging even as enterprise AI adoption is still in relatively early stages — a signal of how steep the capacity curve ahead could be.

What Google Has Said

Google has not publicly disputed the capacity challenge. During its first-quarter earnings call, parent company Alphabet reported that Google Cloud revenue climbed to $20 billion. However, Chief Executive Officer Sundar Pichai acknowledged that computing capacity limitations had prevented even stronger growth and had contributed to a notable increase in the cloud division's order backlog — a rare admission that demand is running well ahead of supply.

Why This Matters

The reported restrictions on Meta's Gemini access illustrate how AI infrastructure shortages are fast becoming a competitive differentiator — and a liability. Companies unable to secure sufficient compute face delays in product development, a risk that grows as AI is embedded deeper into core business functions.

For Google, the episode also surfaces a tension: as a supplier of AI services, capacity rationing could push large customers toward rival clouds or in-house infrastructure, accelerating the very diversification Google's cloud business seeks to prevent. With Microsoft Azure, Amazon Web Services, and a clutch of specialised AI infrastructure providers all vying for the same enterprise budgets, the stakes of a capacity miss are high on both sides of the transaction.

How Google allocates its constrained compute — and how quickly it can expand supply — will be a defining question for the AI cloud market through the rest of 2025.