Meta, SpaceX처럼 여분의 AI 컴퓨팅을 현금화하려 한다

Meta has spent billions of dollars developing artificial intelligence and building out data centers to support it. Now, the company appears poised to deploy that infrastructure for a more immediate profit stream. On Wednesday, Bloomberg reported that Meta is developing plans for a cloud infrastructure business that would sell access to both AI compute power and models. This strategic pivot would position Meta directly against major cloud providers such as Amazon Web Services, Google Cloud, and Microsoft Azure.

Meta’s decision to monetize excess compute capacity follows a similar announcement from SpaceX via xAI just weeks earlier. In early May, SpaceX struck a deal with Anthropic to purchase all available compute capacity at its Colossus 1 data center. Since then, SpaceX has executed comparable leasing agreements with Google and Reflection AI. Meta’s parallel move signals a broader industry shift: the victors in the AI race may ultimately be determined not by who builds the most advanced models, but by who controls the underlying data center infrastructure.

This strategy hinges on sustained demand for compute and the continued valuation of physical data centers. However, market skeptics have cautioned that the aggressive buildout of AI infrastructure risks creating a bubble heavily dependent on rapidly depreciating semiconductor chips. Others have questioned whether AI firms can generate sufficient end-user revenue to justify their trillion-dollar capital commitments.

Despite these concerns, Meta continues to allocate substantial capital toward AI infrastructure. As of the end of the first quarter, the company had committed to spending $182.9 billion on AI infrastructure over the coming years, funding massive ongoing projects in Louisiana and Ohio. The Ohio facility, which Mark Zuckerberg stated would be the size of Manhattan, is scheduled to come online later this year.

Unlike competitors such as Google and OpenAI, Meta has not yet experienced significant external demand for its proprietary AI models and services. The company does not disclose separate revenue figures for Meta AI or Llama, its open-weight AI model family, in its financial reports. Publicly, executives have primarily highlighted the internal enterprise applications of AI, suggesting that Meta’s AI initiatives do not currently constitute a material, standalone revenue stream.

To recoup a portion of its massive capital expenditure, Meta may adopt CoreWeave’s business model by selling access to “raw” compute capacity, according to Bloomberg. The report also indicates that Meta is considering emulating AWS’s approach by offering access to various AI models—including its recently launched closed-weight model, Muse Spark—hosted directly on its AI infrastructure. This new commercial venture will operate under the initiative reportedly named Meta Compute. The project is being spearheaded by Santosh Janardhan, head of infrastructure; Daniel Gross, leader of Meta Superintelligence Labs; and Dina Powell McCormick, president.

The report validates Zuckerberg’s May comments confirming that a Meta cloud computing division is “definitely on the table” as a mechanism to generate returns on the company’s heavy investments in its AI “superintelligence” strategy. TechCrunch has contacted Meta for comment.

When you purchase through links in our articles, we may earn a small commission. This does not affect our editorial independence.

Rebecca Bellan is a senior reporter at TechCrunch covering the business, policy, and emerging trends shaping artificial intelligence. Her work has also appeared in Forbes, Bloomberg, The Atlantic, The Daily Beast, and other publications. She can be contacted or verified via email at rebecca.bellan@techcrunch.com or through encrypted message at rebeccabellan.491 on Signal. View Bio.