In a sign that the artificial intelligence boom is running into real-world limits, Google has reportedly placed a ceiling on how much of its Gemini AI technology Meta can use. The Financial Times reports that Meta's demand for computing power outstripped what Google Cloud could supply, forcing the social media giant to ration its AI usage and delay some internal projects.
What Happened
According to the report, Google told Meta around March that it could not meet the full Gemini capacity Meta wanted. Meta's unusually heavy usage meant it felt the squeeze more acutely than other customers, leading to tighter controls on so-called tokens—the units that measure how much text an AI model processes. The constraints reportedly slowed some of Meta's in-house AI development work.
The situation highlights a growing bottleneck in the AI industry: the physical infrastructure needed to train and run advanced models. These systems require vast numbers of specialized chips, like graphics processing units (GPUs), and power-hungry data centers. Adding that capacity takes time and billions of dollars, even for the world's largest tech companies.
Google Cloud's Own Constraints
Alphabet, Google's parent company, has been describing the same constraint in its own financial results. In the first quarter ended March, Google Cloud revenue reached $20 billion. CEO Sundar Pichai said limited compute capacity held growth back, and the cloud unit's backlog—signed deals it hasn't been able to deliver yet—nearly doubled quarter-on-quarter.
This backlog is a key metric for investors. When a cloud provider is short on compute, demand doesn't show up cleanly in revenue. Instead, customers sign contracts and wait, and those commitments pile into backlog because the cloud firm can't deliver (and bill for) enough GPU-heavy capacity fast enough. That makes backlog a useful clue for how future growth could look. If Google brings new capacity online, cloud revenue can re-accelerate simply by turning already-queued work into delivered services. But until then, capped usage—like Meta's reported Gemini limits—means near-term cloud growth can look artificially constrained even if customer appetite stays strong.
The broader AI infrastructure crunch has also been evident in other parts of the supply chain. Chipmakers like SK Hynix have gained an edge as demand for memory chips used in AI systems remains robust, while companies like Thomson Reuters have seen evidence that legal AI demand is real, even as pricing pressure builds.
What It Means for Investors
For everyday investors, this story is a reminder that the AI boom is not just about software and algorithms—it's also about physical infrastructure that takes time and money to build. The compute shortage means that even the biggest players face limits, and those limits can show up in financial results in unexpected ways.
Google Cloud's nearly doubled backlog is a double-edged sword. On one hand, it signals strong demand that could translate into future revenue once capacity comes online. On the other hand, it means current revenue growth may understate the true level of customer interest. Investors watching Alphabet's cloud business should pay attention to capacity expansion plans and backlog trends, not just reported revenue.
For Meta, the compute constraints mean its AI ambitions may be delayed, which could affect everything from its advertising algorithms to its metaverse projects. The company has been investing heavily in AI, and any slowdown could impact its competitive position.
The broader lesson is that the AI industry's growth is not guaranteed to be smooth. Infrastructure bottlenecks, whether in chips, data centers, or cloud capacity, can create friction even for the most well-funded players. As the EU targets cloud giants like AWS and Azure with new regulations (threatening their lock-in tactics), the landscape could become even more complex.
For now, the Google-Meta episode serves as a concrete example of how the AI boom's infrastructure constraints are playing out in real time. Investors should watch for similar stories from other cloud providers and chipmakers, as they will offer clues about the pace and shape of AI adoption in the months ahead.
