Google 'limits' Meta's Gemini use, but the limit is that it ran out of capacity
The headline reads like a competitive squeeze. The FT story underneath is duller and stranger: Google simply couldn't sell Meta as much Gemini capacity as Meta wanted to buy, and Meta's own staff are now told to ration tokens.
Around March, Google told Meta it couldn't supply the full Gemini capacity Meta had tried to purchase, delaying some of Meta's internal AI work. Meta then asked staff to be more efficient with tokens. Other Google Cloud clients were squeezed too.
The thread's flip: this isn't Google restricting a rival's access, it's a compute shortage dressed as a restriction. And the buried oddity is Meta, which builds Llama, leaning on a competitor's model in the first place.
The word 'limits' is doing a lot of work
CNBC's writeup of a Financial Times report says Google has "limited" Meta's use of Gemini. The substance is narrower than the verb suggests. Around March, Google told Meta it could not meet the full Gemini capacity Meta had sought to purchase, and the shortfall disrupted and delayed some of Meta's internal AI projects. Meta, in response, pushed staff to be more efficient with AI tokens — the units that meter model usage. Other Google Cloud customers were affected as well, the FT says, just less than Meta, whose demand was unusually high.
The backdrop is a sold-out cloud. Google has said compute constraints are capping its growth, with the cloud unit's order backlog nearly doubling quarter on quarter. This is a supplier turning away revenue because it has no spare GPUs, not a platform fencing off a competitor.
The thread corrects the headline before the article does
The top response flagged the framing within minutes:
"This seems to be a bit of a misleading headline… the story here ostensibly seems to be about capacity constraints, not any limitation on what models or capabilities Google is giving Meta access to." — HarHarVeryFunny
An AI author and longtime practitioner agreed on the headline and used the episode to make the opposite of a bubble argument:
"Misleading title on HN but an interesting article, a reminder of why the hyper scalers are investing heavily in infrastructure." — mark_l_watson, author and AI practitioner with 55 patents
The stranger thread is why Meta is buying so much Gemini at all. Meta ships its own Llama models, which made the situation read, to one product manager, like a category error:
"Meta builds its own models. How similar is this to a story with the headline 'OpenAI limits Anthropic's use of its ChatGPT AI models.'?" — a product manager in the thread
Guesses at the actual use case landed on things Llama is weak at and Gemini is cheap at: image and video understanding at scale, content moderation, and creative generation for the AI-made Instagram accounts now pulling millions of followers in weeks. As an engineer noting Gemini Flash's cost-efficiency put it, the Flash series is "quite cost effective" for image and video understanding — not the coding work where Gemini trails.
The quiet signal is about leverage, not access. A frontier lab can now ration its best customers purely because demand outruns silicon, and one of the largest AI companies on earth is renting inference from a direct rival to cover the gaps in its own model lineup. The bottleneck isn't strategy. It's chips.
Get breakdowns like this every day
The top Hacker News stories, unpacked — by email, Telegram, or RSS.
The KIDS Act uses a negligence standard to make age-checking the safe default for every site
The bill never says 'verify everyone's age.' It says platforms are liable if they 'knew or should have known' a user was a minor — and the cheapest way to never be wrong is to check everyone.