Meta is reportedly looking to rent out some of its data center capacity to AI companies, which could help fund its largest infrastructure buildout to date.
The social media giant is exploring several options for selling access to this capacity, including setting up its own cloud division, offering raw compute to select buyers, or making capacity available to companies running on Meta’s existing AI infrastructure, such as Muse Spark.
If Meta sells raw capacity, it would be competing with CoreWeave, Nscale, and other neocloud providers that have built data centers specifically for AI training and inference. Meta announced a long-term agreement with one of these providers, Nebius, in March, under which it could spend up to $27 billion on compute capacity.
That may not be necessary if Meta now has more capacity than it needs for its own AI ambitions. The Meta AI app has been growing in installs, but still sits far behind ChatGPT, Google Gemini, and Claude in overall usage. In the enterprise market, Meta is almost nonexistent, with its AI mostly used to optimize its own ad network.
Meta pulling a SpaceX and joining the neoclouds
The plan to offer additional capacity to rivals would mimic recent deals by SpaceX, including a $1.25 billion-a-month deal with Anthropic and a $920 million-a-month deal with Google, both for access to excess capacity at its Colossus 1 data center.
Like SpaceX, Meta’s data centers are optimized for AI deployments, which could make them attractive to the big names in the AI world.
While the AI buildout is necessary due to the high compute needs of training and inference, there is no guarantee that the hundreds of billions of dollars set aside for infrastructure will all be needed internally. Meta alone spent $72 billion on capital expenditures, mostly AI-related, in 2025, with a planned increase to $125-$145 billion in 2026.
That is why SpaceX and Meta are pursuing deals to sell excess capacity, potentially offsetting some of the cost while demand remains high. Even Google, which has operated its own cloud computing business for over a decade, has struggled to meet demand from customers and its own AI projects, with a recent report suggesting it has limited Meta’s access to Gemini capacity.
More about data centers
- Stargate Norway: OpenAI’s First AI Data Center in Europe
- AI Data Centers’ Soaring Energy Use: Who Pays for Higher Utilities Costs?
- China’s Submerged AI Data Center Could ‘Influence Global Sustainable Computing’
- Google to Power Data Centers With Nuclear Energy by 2030 in First-Of-A-Kind’ Agreement
Another non-social business venture for Meta
Meta has been looking for new ways to generate income outside of ads, which have been its main source of revenue since its inception. Its first major attempt was the metaverse in 2021, but it failed to catch on. After two years of heavy investment, Meta shifted more resources toward AI and smart glasses.
AI has not yet paid for itself, although Meta has said its ad business is seeing strong performance improvements from AI deployment. It hopes to become a contender alongside OpenAI, Anthropic, and Google for consumers using AI for search, which could become a lucrative market once ads are properly integrated.
Meta is also looking at other markets, including a prediction platform in the style of Polymarket and Kalshi, although potentially without real-money spending.
For now, Meta’s reported plans look less like a full cloud challenge to Amazon, Microsoft, or Google and more like a way to make expensive AI infrastructure work harder. But if demand for compute stays high, unused capacity could become more than a cost problem. It could become Meta’s next business experiment.
Related reading: Meta has also recently launched paid subscriptions for WhatsApp, Instagram, and Facebook, offering users more customization features on their accounts.