Google has placed limits on Meta's use of its Gemini AI models, after Google Cloud was unable to fully supply the compute capacity Meta requested. The move underscores how surging AI demand has turned compute scarcity into a pressing industry-wide constraint.
June 2026 · The Compute Squeeze
Google Caps Meta's Use of Gemini — Because It Can't Spare the Compute
Google notified Meta around March 2026 that it could not meet its requested compute capacity, capping Meta's internal use of Gemini. The limit, still in place, has delayed some of Meta's AI projects — a sign that even a $2T giant can't secure all the compute it needs.
Mar 2026
Cap notified to Meta — and still ongoing
$2T
Meta's scale — yet still short on compute
$115–145B
Meta's planned 2026 AI capex
Spending billions — still renting compute
Meta's 2026 AI capital expenditure, in $billions (each block = $25B)
Even at this spend, Meta still depends on external providers — and ran into Google's compute wall.
HOW THE BOTTLENECK CASCADED
Avocado delayed
Meta's frontier model slips, at times trailing Gemini 3.0
→
Leans on Gemini
Meta uses Gemini for internal AI work
→
Google caps use
Can't supply requested compute capacity
→
Tokens rationed
Meta tells staff to use AI tokens efficiently; projects delayed
Compute is the real bottleneck of AI
Google Cloud itself faces capacity constraints and a backlog amid surging demand — and similar limits are spreading across the industry.
The Meta cap is a business-to-business arrangement
Separate from Gemini's consumer "compute-based usage limits"
Hyperscaler capacity allocation now shapes the pace of AI development
Continue reading The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.
Already purchased? Sign in ✓ Signed in — this article isn’t included in your current plan.Unlocking the full article…