Google limits Meta's use of Gemini models amid compute crunch, touts TPU cost edge

Reuters, citing the Financial Times, reported June 28 that Google restricted Meta's access to its Gemini models after Meta requested more compute than Google could supply. The episode crystallizes the week's second major theme — a hard ceiling on AI compute — and shows providers rationing capacity even to large paying customers rather than ceding it to a direct rival.

The mechanics matter: Google sells Gemini access to enterprises while simultaneously competing with them, so throttling Meta is both a capacity decision and a competitive one. Alphabet's leverage here is vertical integration. A CNBC piece the prior day detailed how Alphabet leans on its custom TPUs, with Pichai crediting them for a 78% reduction in Gemini serving unit costs across 2025 — a structural margin and pricing advantage rivals dependent on merchant NVIDIA silicon lack.

Competitively, the story sits alongside Amazon and Apple raising prices on chip-cost pressure and xAI renting out idle Colossus GPUs, painting an industry where compute scarcity, not model quality, is the binding constraint. An HN thread (146 pts) on Google limiting Meta's Gemini use highlighted the growing tension between AI providers over competitive access.

Watch whether Meta accelerates its own silicon and capacity buildout in response — it already runs massive data centers for AI workloads — and whether other Gemini enterprise customers face similar limits as Google prioritizes its own products. The throttling also raises a strategic question for any enterprise standardizing on a frontier provider that is simultaneously a competitor.