Google Capped Meta's Gemini Access — Here's How the AI Compute Crunch Unfolded

In late June 2026, a Financial Times report revealed a crucial moment in the AI industry's infrastructure story: Google had placed a cap on Meta's use of its Gemini AI models because Alphabet simply did not have enough computing capacity to meet the social media giant's demand . The restrictions, implemented around March 2026, didn't just limit Meta's access — they disrupted internal AI projects, forced a strategic pivot, and signaled a broader reckoning with AI compute scarcity that would reshape how even consumers pay for AI .

What actually happened

Around March 2026, Google informed Meta that it could not supply the full volume of Gemini API capacity Meta wanted to purchase . Google enforced a cap on Meta's Gemini access — one of its largest customers — due to acute AI compute shortages . Multiple Google cloud customers faced similar rationing, but Meta, with its exceptionally high demand, was hit hardest .

How the shortage hit Meta's internal projects

The capacity shortfall "disrupted and delayed some of Meta's internal AI projects," according to the Financial Times . Sources told the FT that the restrictions, which remained in place as of late June 2026, forced Meta to recalibrate its AI roadmap significantly .

Meta's two-pronged strategic response

• Efficiency mandates. Meta asked its staff to use AI tokens more conservatively — doing more with less compute per query . The company explicitly told employees to optimize token usage across its internal systems .
• Accelerated proprietary model development. With less access to a rival's infrastructure, Meta has had to push harder on developing its own in-house AI models to reduce dependency on Google's stack .

This dual strategy — use less from the external provider while building more internally — mirrors how many large enterprises are responding to AI supply constraints.

The broader industry context: a systemic compute crunch

The Meta-Gemini cap is the most prominent signal of a systemic compute crunch across the AI industry . The hardware and energy infrastructure required to train and serve frontier AI models simply hasn't kept pace with surging demand, creating bottlenecks affecting even the largest technology companies .

How Google's consumer subscription model changed in response

At Google I/O 2026 (May 19–20), the company itself responded to the capacity tension by overhauling its Gemini subscription plans. Google moved from fixed daily prompt limits to a compute-based usage model measured by actual processing consumption .

Under the new system, usage is metered based on three factors :

• Prompt complexity (simple text queries cost less than complex video or coding prompts)
• Features used during a session
• Chat length (longer, multi-step conversations consume more)

Limits refresh every five hours until a weekly cap is reached . If a user hits their cap, the system can fall back to smaller models rather than cutting off access entirely .

Google introduced a restructured three-tier subscription lineup :

Tier	Price	Key Features
Google AI Plus	$7.99/month	Baseline compute-based limits
Google AI Pro	$19.99/month	Increased limits, YouTube Premium Lite in select countries
Google AI Ultra	$99.99/month	5x higher limits than Pro, Gemini 3.5 Flash, Google Antigravity access, 20TB cloud storage, YouTube Premium

The previous top-tier Ultra plan was reduced from approximately $250 to $200 per month and includes 20x the compute limits of the Pro plan .

This shift mirrors a broader industry trend toward consumption-based AI billing, similar to what ChatGPT and Claude have already adopted .

Why this matters

The Meta-Gemini cap and Google's subscription overhaul are two sides of the same coin: the AI industry is hitting hard physical limits on compute infrastructure. Even the biggest players — with the deepest pockets — cannot simply buy their way past the chip, power, and data-center bottlenecks . The result is strategic rationing for enterprise customers and more nuanced metering for consumers, with implications for every company building on top of these models. Meta's forced pivot to proprietary models may also accelerate the trend of large tech companies insourcing their AI capabilities rather than relying on rivals' platforms.

What the sources say

The core story was first broken by the Financial Times on June 28, 2026, citing three unnamed sources familiar with the matter . It was immediately corroborated by Bloomberg, CNBC, Benzinga, and Reuters . The Google I/O 2026 subscription changes were reported by multiple outlets including Mashable, Times of India, and Business Standard .

Confidence level: High. The core facts — Google capping Meta in March due to compute shortages, disruption to Meta's projects, Meta urging token efficiency, and the Google I/O shift to compute-based billing — are well-supported across multiple credible outlets. No contradictory reporting has been found.