As the artificial intelligence (AI) infrastructure shortage becomes a reality, Google has reportedly restricted Meta’s usage of AI models. This computing bottleneck is now affecting even global tech giants, preventing them from utilizing AI models to their desired extent due to skyrocketing AI demand.
The Financial Times (FT) reported on June 28 that Google informed Meta in March it couldn’t provide all the computing capacity they requested, resulting in limitations on Meta’s use of the Gemini AI model.
While Meta has its own AI model, Llama, it doesn’t operate its own cloud services. Instead, it has been using competitive AI models like Google’s Gemini and Anthropic’s Claude for internal operations.
However, as AI service demand surged, the scarcity of computing resources exposed the vulnerabilities of heavily relying on external AI models. This move has reportedly disrupted some of Meta’s key AI projects.
In response, Meta has advised its employees to manage their AI token usage more efficiently.
Other Google AI clients have also received similar usage restrictions. The FT noted that the impact on Meta, which heavily relies on Gemini, has been particularly significant.
AI Companies Now Compete on Efficiency As Computing Resource Scarcity Becomes Reality
Operating AI models requires extensive computing infrastructure, including graphic processing units (GPUs), servers, and data centers. Despite major investments in AI data centers by global tech firms, the rapid increase in AI demand has outpaced supply, leading to a shortage of computing resources.
The FT assessed that Google’s restriction on model access for major clients illustrates the infrastructure pressures and bottlenecks emerging across the AI industry.
As AI computing resource scarcity becomes apparent, AI companies are adapting their service operations.
This month, Google signed a multi-year cloud service contract with SpaceX, which has built a massive AI data center called Colossus. From October through June 2029, Google will pay 920 million USD monthly for computing resources.
Following suit, Anthropic has entered a similar agreement with SpaceX and decided to implement a pay-per-use model for its previously subscription-based AI services.
Companies that have actively utilized AI are shifting their strategy from quantity to efficiency.
Andrew Bosworth, Meta’s Chief Technology Officer (CTO), emphasized in an April memo to employees that using AI tools shouldn’t be an end in itself, stressing that high token usage doesn’t necessarily lead to results.
Industry analysts suggest that beyond model performance, securing computing resources and optimizing token usage efficiency are becoming new competitive advantages in the AI sector. The focus is shifting from how much AI is used to how effectively it’s utilized, which is now a key factor in determining a company’s competitiveness.
