Google reportedly limited Meta's Gemini access over limited AI compute

Even Google is reportedly short on Gemini capacity for major AI customers as it could not meet Meta's high demand for Gemini.

Google is reportedly limiting Meta's use of its Gemini AI models after Meta tried buying more computing capacity than even Google could supply.

According to the Financial Times, Google told Meta in March that it could not provide the full Gemini capacity that Meta had requested. This shortfall even disrupted and delayed some of Meta's internal projects. Due to this, Meta even told its employees internally to use AI tokens more efficiently.

Meta wasn't the only one to get hit by this sudden refusal by Google; even other customers were affected. But Meta was hit harder because of its unusually high demand for Google's models.

The move from Google makes it evident that companies all over are in limited supply of both infrastructure and compute. Alphabet said in April that Google Cloud revenue grew 63% year-over-year to $20 billion in the first quarter, helped by enterprise AI infrastructure and AI solutions.

In pursuit of more compute, Meta had earlier signed a multi-billion-dollar AWS agreement as well as a large AMD GPU deal for AI data centers.

But the crunch would be short-lived as both Meta and Google have also ramped up infrastructure investments heavily. Meta said in November that it was committing more than $600 billion in the U.S. by 2028 for AI technology, infrastructure, and workforce expansion. In the first quarter of this year, Meta also raised its expected capital expenditure for 2026 to a range of $125 billion to $145 billion, citing higher component pricing and additional data center costs for future capacity.

However, this doesn't make the company immune to the current dependence on outside suppliers.

Meta has also spent many years promoting Llama as an open-weight alternative to closed models from Google, OpenAI, and Anthropic. But if the reported reliance on Google's Gemini models is severe enough for internal work to get impacted, then it looks like even frontier labs and Big Tech aren't fully self-sufficient.

Source: Financial Times