Google Throttled Meta's Gemini AI Access in March Over Compute Shortage
- Google told Meta in March 2026 it could not provide all the Gemini compute capacity Meta requested, delaying multiple internal AI projects [1]
- Google Cloud carries a nearly $460 billion unmet demand backlog despite quarterly revenue exceeding $20 billion, up 63% year-over-year [3]
- Meta instructed employees to use AI tokens more sparingly and began shifting workloads to its internal Muse Spark model [3]
- Google plans $180–190 billion in 2026 infrastructure spending and is leasing additional capacity from SpaceX at roughly $920 million per month [3]
- The original Meta-Google deal, signed in August 2025, was valued at over $10 billion across six years [1]
Google informed Meta in March 2026 that it could not fulfill the full computing capacity Meta had sought for access to its Gemini AI models, according to the Financial Times, citing three people familiar with the matter [1]. The restriction delayed multiple internal AI projects at Meta and prompted the company to direct employees to use AI tokens — the units measuring AI model usage — more sparingly [2].
Meta is among Google's largest Gemini customers under a deal signed in August 2025 valued at more than $10 billion over six years [1]. Other Google Cloud customers also experienced limited access, though to a lesser degree, according to the reports [2]. Neither Google nor Meta responded to requests for comment.
The throttling episode is the most concrete public evidence yet of compute demand outstripping supply at the hyperscaler level. Google CEO Sundar Pichai acknowledged the constraint on a recent earnings call, stating that Cloud revenue "would have been higher if we were able to meet demand" [3]. Google Cloud's quarterly revenue exceeded $20 billion — up 63% year-over-year — but the unit's unmet demand backlog nearly doubled quarter-over-quarter to approximately $460 billion [3].
The Capacity Crunch
The root cause is straightforward: demand for GPU-backed inference and training capacity is growing faster than Google can deploy data center infrastructure. Despite planning $180–190 billion in capital expenditure for 2026 — a figure that would make it the largest single-year infrastructure spend in corporate history — Google cannot close the gap quickly enough [3].
To supplement its own build-out, Google has turned to unconventional capacity sources. The company is leasing AI data center capacity from SpaceX at approximately $920 million per month, and has entered similar arrangements with xAI [3]. Anthropic has also entered a comparable capacity deal with Google, according to reporting by The Bridge Chronicle [3].
Meta's Response
Meta has moved to reduce its dependence on external AI providers in the wake of the restrictions. The company began shifting affected workloads to Muse Spark, an internally developed model, according to multiple reports [3]. Employees were also directed to improve token usage efficiency across projects [2].
The pivot comes as Meta pursues its own massive infrastructure build-out, with plans to invest up to $135 billion in AI infrastructure [3]. Meta cut approximately 8,000 jobs as part of a broader reallocation toward AI spending [3]. Despite those investments, Meta's reliance on Google's Gemini models for coding tools, customer service, advertising optimization, and content moderation made the capacity restriction operationally significant [2].
Market Context
Alphabet (GOOGL) shares closed at $352.55 on June 29, up 4.5% on the day and roughly 12.7% year-to-date, giving the company a market capitalization of approximately $4.26 trillion [4]. Meta (META) shares traded at $564.68, up 2.6% on the session but down 14.4% year-to-date at a $1.43 trillion market cap [4].
The divergent stock trajectories reflect broader investor sentiment: Google's cloud business is capacity-constrained but growing at 63% annually, while Meta's share price has been weighed down by heavy AI capital commitments and competitive pressures. Both companies, however, are among the largest capex spenders globally — a group that now includes Microsoft, Amazon, and Oracle — all racing to deploy GPU clusters at scale.
Implications for the Cloud Market
The Google-Meta episode illustrates a structural challenge for the hyperscale cloud model: even the largest customers with multi-billion-dollar contracts cannot guarantee access to compute capacity when demand spikes industry-wide. The $460 billion backlog figure at Google Cloud alone suggests the constraint is not a temporary bottleneck but a sustained supply-demand mismatch [3].
For enterprise customers without Meta's purchasing power, the implications are starker. If a $10 billion-plus contract does not secure priority access, smaller customers will need to diversify across providers, build private capacity, or accept queue times. The episode is likely to accelerate interest in neocloud providers like CoreWeave and Lambda, as well as sovereign and on-premise AI infrastructure deployments.
Google's reliance on third-party capacity from SpaceX and xAI — companies that are themselves heavy compute consumers — adds another layer of complexity. The leasing arrangements suggest Google views near-term capacity shortfalls as severe enough to justify premium economics for rented infrastructure [3].
Companies mentioned
Alphabet Inc. provides a diverse range of products and digital platforms to consumers across multiple global regions, including North and South America, Europe, the Middle East, Africa, and the Asia-Pacific. The company…
Meta Platforms Inc., which operated as Facebook, Inc. until its October 2021 rebranding, is a technology enterprise focused on developing innovative products that empower people globally to connect and share with their …