Google had to cap Meta’s access to its Gemini AI model after Meta burned through more computing capacity than Google could supply, according to sources familiar with the matter who spoke to Engadget. The situation puts a spotlight on a growing problem across the entire AI industry: there simply is not enough computing power to go around, even for the biggest players.
What makes this story unusual is who’s involved. Meta is not some scrappy startup. It’s one of the wealthiest tech companies in the world, with billions committed to AI infrastructure. And yet it was still leaning heavily on Google’s Gemini for key parts of its business, because Gemini outperformed Meta’s own Llama open-source models for certain tasks.
Meta uses Gemini for a range of functions, including:
- Customer service and advertiser chatbots
- Internal coding tools
- Harmful content detection and takedowns
- Scam detection
Google reportedly flagged the capacity issue to Meta back in March. In response, Meta had to ask employees to use tokens more efficiently, essentially asking people to get more done with less. The company also uses Anthropic’s Claude for similar purposes, so Gemini is not its only external AI dependency.
This matters beyond just Meta and Google. It shows that the AI supply chain is under serious strain at every level. Google itself recently agreed to pay SpaceX $920 million a month to use xAI’s data centers because it needed extra computing capacity to run Gemini Enterprise. A company that makes AI is now paying a competitor’s infrastructure to keep up with demand for its own product.
Meta does not run its own cloud business, which makes it more dependent on outside providers than rivals like Microsoft or Amazon. The company has pledged $600 billion in cloud computing investments over the next two years, but that buildout takes time. In the meantime, it has to work with what’s available, and right now, not much is.
The broader economics here are worth noting. Token prices have been rising, and some companies are already pulling back on how much AI they use because of cost. That includes, awkwardly, the AI companies themselves. Analysts have pointed out that revenue from AI still represents a small fraction of what companies are spending to run it. The boom is real, but so far, profitability is not.
For anyone watching the AI industry closely, this episode is a useful reality check. Demand for AI computing has grown faster than anyone built capacity to handle it. Until that gap closes, stories like this one are going to keep happening.
