Meta sells cloud AI capacity: a strategic shift

What happened?

Meta is designing a cloud computing business to sell excess capacity from its AI data centers to third parties, according to Bloomberg (source: Xataka). The company is considering two approaches: offering access to models hosted on its infrastructure (such as its upcoming Muse Spark model) or directly renting out computing power in the style of neoclouds (CoreWeave, Nebius). This move represents a significant strategic shift for a company that has so far been primarily a massive buyer of cloud infrastructure, not a seller. Zuckerberg had hinted at this move during the May 2026 shareholder meeting, noting that competing in the cloud was 'definitely on the table' and that companies asked him 'almost every week' if they could buy compute from him. Now that trial balloon has turned into a concrete business plan.

Why is it important?

Meta built one of the world's largest computing infrastructures between 2023 and 2026 almost single-handedly, without partners or external customers to monetize it. 98% of its revenue still comes from advertising. Turning those data centers into a product that is sold, rather than just a cost to be borne, is the first public sign that Meta needs another revenue source to justify the expenditure. According to analyst estimates, Meta's capital expenditure on AI infrastructure exceeded $35 billion in 2025, and is expected to reach $45 billion in 2026. Without cloud revenue, this spending is difficult to amortize. Comparatively, AWS, Azure, and Google Cloud generate hundreds of billions in annual revenue. Meta now seeks a slice of that pie, though initially it would focus on customers looking for specialized AI capacity, a niche where neoclouds like CoreWeave (which billed $2 billion in 2025) have shown there is demand.

Consequences and context

This move puts Meta in direct competition with cloud providers like AWS, Google Cloud, and Azure, though initially it would focus on customers seeking specialized AI capacity. It also reinforces the trend of big tech companies monetizing their AI infrastructure. This is not the first time a social media company has tried to diversify into the cloud: in 2016, Twitter considered selling its data infrastructure but never materialized it. Meta, however, has the scale and the need. Meta's infrastructure includes over 600,000 GPUs (as of 2025 data), and it plans to reach one million by 2027. This excess capacity, once considered a sunk cost, is now seen as a revenue-generating asset. However, Meta faces challenges: it lacks the enterprise sales experience of its competitors and will have to build a sales and support team from scratch. Moreover, price competition is fierce: AWS and Azure offer volume discounts that Meta will struggle to match initially.

What readers should know

• The service would include access to Muse Spark, Meta's own model still without a release date for external developers. Announced in 2025, this model is designed for multimodal reasoning tasks and could be a differentiator against OpenAI or Google models.
• Also under consideration is renting out raw GPU power, the business model of CoreWeave and Nebius. CoreWeave, which started as a cryptocurrency miner, has scaled to become a GPU cloud provider valued at $19 billion. Nebius, meanwhile, is a spin-off from Yandex that has raised $2 billion in investments.
• Meta still relies on advertising for 98% of its revenue; this move aims to diversify income. In 2025, Meta's advertising revenue was $160 billion, while non-advertising revenue (mainly Reality Labs) barely reached $2 billion. A cloud business could add between $5 billion and $10 billion in annual revenue by 2028, according to Bernstein estimates.
• Meta's infrastructure is one of the largest in the world, built between 2023 and 2026. It includes data centers in the US, Europe, and Asia, with expansion plans in Latin America. Meta has opted for custom data center designs optimized for AI training, giving it an energy efficiency advantage over generic data centers.

In summary, Meta is taking a bold step to monetize its AI investment, but success will depend on its ability to execute a competitive cloud business in a market dominated by giants. The next 12-18 months will be critical to see if this move becomes a new revenue stream or an expensive experiment.