Meta has been forced to ask employees to use AI tokens more carefully after Google capped access to its models, exposing how badly the industry’s infrastructure is lagging behind its own ambitions
The AI industry has spent two years making bold promises about what its technology can do. It is now running into a more fundamental problem: there is not enough computing power to go around, and even the biggest companies in the world are feeling the squeeze.
What Happened Between Google and Meta
When Meta approached Google in March this year seeking additional compute capacity for its internal AI projects, Google said no. According to a Financial Times report published on June 28, Google imposed limits on Meta’s use of its Gemini AI models, telling the social media giant it could not fulfil the full capacity Meta had sought to purchase. Several other Google customers were also affected, but Meta bore the brunt of the restrictions given the scale of its demand.
The development is not entirely surprising. Google CEO Sundar Pichai flagged during the company’s last earnings call that compute constraints were already preventing higher growth in Google Cloud and had contributed to a backlog that nearly doubled quarter on quarter. Google Cloud revenue reached $20 billion in the first quarter of this year.
The Ripple Effects Inside Meta
The capacity crunch has had direct consequences for how Meta manages AI usage internally. Earlier this year, the company had encouraged employees to use AI tools as extensively as possible, a trend the industry had started calling tokenmaxxing. Meta even tied employee performance evaluations to AI tool usage. That posture has now quietly reversed, with staff being encouraged to use AI tokens more efficiently.
A Wider Industry Problem
Meta is not alone in navigating these constraints. Microsoft’s Experiences and Devices division, which covers Windows, Microsoft 365, Outlook, Teams, and Surface, was instructed to wind down its use of Anthropic’s Claude Code by the end of June. While part of the reason involves steering developers toward Microsoft’s own Copilot CLI, financial considerations reportedly played a role too.
The pattern is consistent across the industry. Billions are being spent on chips and data centres, but demand for AI compute is growing faster than the infrastructure being built to support it.
