Computing power emergency: Google quietly imposes Gemini usage cap on Meta.

The supply-demand conflict in AI infrastructure is intensifying among the world’s leading technology companies. According to sources, Google informed Meta around March this year that it could not meet Meta’s full Gemini computing capacity requirements, and imposed usage limits on this social-media giant—showing that even the world’s largest AI service provider is struggling to keep up with the surge in computing demand.

As reported by the British Financial Times, these restrictions have not been lifted to date, causing disruptions and delays to multiple internal AI projects at Meta. As a result, Meta has asked employees to improve the efficiency of AI computing power usage and to practice careful budgeting of AI tokens internally. Both Google and Meta declined to comment on the matter.

This situation has forced Google to speed up its capacity expansion. Earlier this month, Google signed a monthly computing power leasing agreement worth $920 million with SpaceX, which is under Elon Musk. Speaking during the company’s Q1 earnings call, Google CEO Sundar Pichai acknowledged, “Recently, we are indeed facing constraints in computing power. If we could meet demand, our cloud business revenue would be higher.”

Meta is not an exception. Multiple sources indicate that other Google enterprise customers are also affected by restrictions to varying degrees, but Meta has been hit the hardest due to its unusually massive demand. The turmoil reflects the explosive growth of AI inference workloads, which has become one of the industry’s biggest challenges.

Computing power bottlenecks continue to mount pressure, and big customers are hit first

Even though major tech companies have poured hundreds of billions of dollars into chips, data centers, and power supply, AI computing power supply still has difficulty keeping pace with the rate of demand growth.

In Q1, Google’s cloud business revenue first surpassed $20 billion. The amount of cloud contracts that have been signed but not yet delivered piled up to nearly double quarter-over-quarter, exceeding $460 billion. Pichai has made it clear that the computing power constraints will continue in the near term.

Against this backdrop, Meta’s impact is particularly pronounced. Sources say that it is precisely the high-intensity demand from large enterprise customers such as Meta that directly drives Google to accelerate its search for external computing power sources. As enterprises deploy chatbots, programming assistants, and AI agents at large scale, inference workloads—i.e., the computing power consumed by models to perform tasks in real-world applications after training—are becoming the industry’s core bottleneck.

Meta’s internal projects are obstructed, accelerating the shift to self-developed models

Meta makes extensive internal use of Gemini, covering platform safety reviews (including identifying scam content and removing harmful information), customer service and advertising-assisted chatbots, as well as some internal workflows and code development, while also using other models such as Anthropic’s Claude alongside it.

According to sources, Meta initially chose Gemini because it performed better than the company’s own self-developed open-source Llama model. However, as computing power constraints have tightened, Meta is accelerating its migration to self-developed models. Multiple sources say that Meta has recently begun prioritizing its newly launched Muse Spark model, which is believed to have performance on par with Gemini, helping to reduce reliance on external models.

Meta CEO Mark Zuckerberg has previously continued to increase investment in AI talent and infrastructure, aiming to build what he calls “personal superintelligence.” Unlike Google, Meta has no cloud business. It is accelerating the construction of its own data center system and has pledged total investment of $600 billion in the United States by 2028.

Google expands via SpaceX, and the industry seeks a breakthrough

Facing computing power pressure, Google signed this month a $920 million monthly computing power leasing agreement with SpaceX to make up for infrastructure gaps. AI lab Anthropic also reached a similar agreement with SpaceX last month.

Google’s use of restrictions against Meta gives the public a rare window into the real pressure faced by the world’s top AI service providers in allocating computing power. At present, the infrastructure bottleneck across the AI industry is spreading from the training side to the inference side, and resolving the supply-demand mismatch still depends on the delivery and rollout of another round of large-scale capital investments.

Risk warning and disclaimer

        There are market risks, and investment requires caution. This article does not constitute personal investment advice, and it does not take into account any special investment objectives, financial situations, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article are suitable for their specific circumstances. Investing based on this article is at your own risk.