Google told Meta it can’t provide all the Gemini capacity Meta wants. Meta’s internal AI projects were delayed. Google’s cloud backlog nearly doubled. And Meta is now telling employees to use AI tokens more efficiently. The AI compute crunch isn’t hypothetical — it’s hitting the biggest companies on Earth.
The Compute Crunch
What the FT Reports
Google told Meta around March 2026 that it could not provide all of the Gemini capacity Meta wanted to purchase. The shortfall disrupted and delayed some of Meta’s internal AI projects.
Meta has been “particularly impacted because of its exceptionally high demand” — but other Google Cloud clients have also been affected. Sundar Pichai acknowledged that compute constraints prevented even higher growth and caused the cloud backlog to nearly double quarter over quarter.
Meta’s response: telling employees to use AI tokens more efficiently — including reducing consumption. The world’s largest social media company, with a $14.3B AI budget, is rationing its AI usage.
The key insight: Meta has $14B+ to spend on AI. Google has $20B/quarter in cloud revenue. And Google still can’t give Meta what it wants. When the buyer has unlimited budget and the seller has $20B in revenue but can’t meet demand, the constraint isn’t money — it’s physical infrastructure. GPUs, memory, power, data centers. The substrate bottleneck is real.
The Structural Read
META RATIONING TOKENS = THE CFO GUIDE IN ACTION
We published the CFO’s Guide to the Token Economy this week. The thesis: tokenmaxxing is over, token arbitraging is the new game. Meta just proved it — telling employees to reduce token consumption. When a $1.4 trillion company rations AI tokens, the economics of inference have become a management problem.
THIS EXPLAINS THE TOKEN SHARE COLLAPSE
Big 3 token share fell from 72% to 33% in one year. Google capping Meta’s Gemini access is part of why — when the premium provider can’t supply enough, customers go to cheaper open-source alternatives. Capacity constraints accelerate the shift to DeepSeek, GLM-5.2, and Llama.
RAMAGEDDON’S ROOT CAUSE CONFIRMED
Why are MacBooks $200 more and Xbox $150 more? Because AI companies are consuming more compute and memory than the world can produce. Google capping Meta proves the same constraint is hitting AI companies themselves, not just consumer electronics. Everyone is competing for the same finite substrate.
The Bottom Line
When Google — with $20 billion in quarterly cloud revenue — can’t meet Meta’s AI demand, the compute crunch is no longer a startup problem. It’s a Big Tech problem. Meta is rationing tokens internally. Google’s backlog doubled. Consumer electronics prices are rising because AI consumed the memory supply. And power companies are IPO’ing at record pace because data centers need electricity nobody planned for. The AI boom’s binding constraint isn’t intelligence. It’s atoms — silicon, memory, electricity, and the physical infrastructure to house them.
Source: Financial Times — June 28, 2026