Google Limits Gemini Usage for Meta Due to 'Computing Power Shortage'

Google has limited the usage of its Gemini artificial intelligence (AI) model for its competitor, Meta Platforms (Meta), the Financial Times (FT) reported on June 28, citing internal sources.

This measure was taken because Meta requested an amount of AI computing power that Google found difficult to accommodate, highlighting that even the world's top AI companies are struggling with infrastructure shortages and bottlenecks.

According to the FT, Google notified Meta in March that it could not provide the full capacity of Gemini that Meta intended to purchase.

This restriction has reportedly caused setbacks for some of Meta's internal AI projects and remains in effect.

To cope with these constraints and control AI costs, Meta is encouraging its employees to use "AI tokens," the unit of AI usage, more efficiently.

While several other Google clients have also been affected by these restrictions, the impact was less severe, and the FT noted that Meta was hit particularly hard due to its exceptionally high demand for Google's models.

According to industry insiders, Meta heavily utilizes Gemini for key tasks such as automating internal security processes, combating online fraud, blocking harmful content, and conducting internal development work.

It appears that Meta decided to use a competitor's AI for its operations because its own models, such as Llama, significantly lag behind Gemini in terms of performance.

Although Meta CEO Mark Zuckerberg has pledged to develop highly advanced AI and is investing unprecedented amounts of money into recruiting talent and expanding infrastructure, he has been unable to bridge the immediate technological gap.

The FT reported that both Google and Meta declined to comment on the report.

This situation clearly illustrates the growing infrastructure pressure within the AI industry.

Even as companies pour tens of billions of dollars into chips, data centers, and electricity, it remains difficult for even the largest tech firms to secure the computing power necessary to run advanced models and AI services.

Google is scrambling to secure additional capacity as demand from major corporate clients like Meta continues to soar.

Earlier this month, Google signed a contract worth $920 million (approximately 1.4164 trillion won) per month to lease computing infrastructure from Elon Musk's SpaceX.

During the first-quarter earnings call in April, Google CEO Sundar Pichai announced that the company's cloud revenue had surpassed $20 billion (approximately 30.8 trillion won) for the first time.

He also stated that the company's cloud backlog—contracts signed but not yet delivered—had nearly doubled from the previous quarter to over $460 billion.

"We are clearly facing computing constraints in the short term," Pichai said. "If we had been able to meet the demand, our cloud revenue would have been even higher."

Demand for AI computing is rising rapidly as companies adopt chatbots, coding assistants, and AI agents across their operations.

Unlike competitors such as Google or Microsoft (MS), Meta does not operate its own cloud business, making it more vulnerable to infrastructure bottlenecks regarding AI development and service operations.

To overcome these limitations, Meta has previously announced plans to invest $600 billion (approximately 923 trillion won) in U.S. AI infrastructure by 2028.

(Photo: Yonhap News)
※ Please note: This article was translated by AI and may contain errors.