An inability to meet Meta $META's appetite for computing resources led Google $GOOGL to restrict the social media company's access to its Gemini AI models, the Financial Times reported.
The restrictions disrupted some of Meta's internal AI projects and prompted the company to tell staff to use AI tokens more efficiently
Bloomberg / Getty Images
An inability to meet Meta $META's appetite for computing resources led Google $GOOGL to restrict the social media company's access to its Gemini AI models, the Financial Times reported.
According to CNBC, as far back as March, Meta learned it would not receive the full allotment of Gemini capacity it had tried to buy. The gap between what Meta needed and what Google could supply set back a number of Meta's internal AI efforts. Other Google customers felt the pinch as well, though none as sharply as Meta, whose outsized appetite for the models made it the most exposed.
Facing the supply crunch, Meta asked employees to be more judicious in how they consumed AI tokens — the basic unit by which AI usage is tracked and metered.
Gemini had become a key tool for Meta in moderating content — handling tasks like flagging harmful posts and rooting out scams — precisely because it outperformed the company's homegrown Llama models for those functions, Bloomberg reported. Muse Spark, Meta's newer proprietary model, has been taking on a larger share of that work as the company tries to bring more of its AI capabilities in-house.
Google and Meta both declined to comment to the Financial Times.
The situation is part of a wider squeeze on AI infrastructure. According to the outlet, Google Cloud brought in $20 billion in revenue during the first quarter ended March, yet Sundar Pichai acknowledged that compute shortages held back what could have been even stronger results — and that unmet demand pushed the division's order backlog to nearly twice what it had been the previous quarter.
Changes Google made to Gemini's usage limits for subscribers earlier this year also pointed to tightening compute constraints. After users complained about hitting caps with only a handful of prompts, Google switched from a prompt-based system to one that weighs request complexity, tool usage, and conversation length when calculating how quickly a user's quota is consumed.
Unlike Google, Meta has no cloud business of its own, and Mark Zuckerberg has positioned AI as the company's central investment focus. As part of a sweeping reorganization described by the outlet, the company announced plans to eliminate roughly 8,000 positions while simultaneously moving 7,000 employees into roles tied to AI work.
Join 500,000+ readers who start their day with Quartz.
By subscribing, you agree to our Terms of Service and Privacy Policy.