Meta is exploring a new revenue stream by selling spare computing power to other AI companies, a shift that suggests the social media giant built more AI infrastructure than it currently needs. The company plans to launch a cloud business offering access to excess capacity, competing with established players like Amazon Web Services (AWS), Microsoft Azure, and Google Cloud. This move comes as Meta grapples with disappointing AI model performance and mounting infrastructure costs.

Why Is Meta Suddenly Selling Computing Power?

Meta's pivot to selling compute capacity reflects a broader challenge facing the company. CEO Mark Zuckerberg committed to spending approximately $600 billion on artificial intelligence (AI) infrastructure through 2028, with a caveat that the company might invest even more if AI progress accelerates. However, Meta's own AI models have underperformed compared to competitors. Following the disappointing launch of Llama 4 last year, the company launched an aggressive hiring campaign, spending tens of millions to recruit AI talent from rivals. In April, Meta unveiled Muse Spark, part of a new family of models developed by Meta Superintelligence Labs.

The gap between Meta's massive infrastructure investment and its model performance has created an opportunity. Rather than let expensive computing hardware sit idle, the company is monetizing the excess capacity. This strategy mirrors what SpaceX is already doing through its xAI division. In May, AI safety company Anthropic signed a deal to use xAI's Memphis Colossus 1 supercomputer, gaining access to more than 300 megawatts of new capacity and over 220,000 Nvidia graphics processing units (GPUs) within a month. Google is also pursuing similar arrangements, agreeing to pay SpaceX $920 million per month for access to compute capacity, including 110,000 Nvidia chips.

What Does This Signal About the AI Infrastructure Market?

Meta's move to sell excess compute capacity has sparked debate about whether the AI infrastructure boom may be cooling. Some analysts interpret the announcement as evidence that the company overestimated demand for its own AI models and overbuilt its infrastructure accordingly. The implications extend beyond Meta. If the company is admitting to excess capacity, industry observers suggest other major tech firms may face similar challenges, potentially leading to reduced or stalled AI spending across the sector.

The semiconductor industry reacted sharply to the news, with chip stocks selling off significantly. This reaction underscores concerns that the AI business model may be broken or at least less profitable than initially anticipated. Companies that have bet heavily on AI infrastructure expansion may need to recalibrate their spending plans.

How Is Meta Approaching Its Broader AI Strategy?

Beyond selling compute power, Meta is pursuing other AI-related ventures. The company nearly acquired Kalshi, a prediction market platform that reached a $22 billion valuation following a $1 billion funding round in May 2026, before Zuckerberg decided to build a competing product internally. Instead of buying Kalshi, Meta instructed its engineering team to develop an application called Arena, which will use play money managed by artificial intelligence systems.

The decision to build rather than buy reflects Meta's broader strategy of developing proprietary AI capabilities. However, the regulatory environment for prediction markets remains complex. The Department of Justice maintains open investigations into alleged insider trading at competing firms, and some states classify prediction market operations as felonies. Meta's choice to use play money rather than real currency is designed to bypass classification as a gambling product, allowing the company to sidestep these regulatory challenges.

Steps Meta Is Taking to Monetize AI Infrastructure

• Cloud Business Launch: Meta is building a cloud infrastructure service to compete with AWS, Azure, and Google Cloud by offering access to spare computing capacity at competitive rates.
• Talent Acquisition: The company is spending tens of millions to recruit top AI researchers and engineers from competitors, strengthening its internal AI development capabilities.
• Internal Product Development: Rather than acquiring AI-related startups, Meta is building competing products like Arena in-house, maintaining control over technology and regulatory compliance.

Meta's infrastructure challenges reflect a broader tension in the AI industry. Companies have invested enormous sums in computing power, betting that demand for AI services would justify the expense. However, if major players like Meta are now selling excess capacity, it suggests the market may not be absorbing compute resources as quickly as anticipated. The coming months will reveal whether other hyperscalers face similar overcapacity issues and whether Meta's cloud business can generate meaningful revenue from its infrastructure investments.