Meta wants to sell surplus data capacity as a cloud service

Meta is exploring whether it can sell computing capacity from its AI data centres to external customers, which would bring the company into a market currently dominated by Amazon Web Services, Microsoft Azure and Google Cloud. Bloomberg reported this on 1 July. It is the first time Meta has publicly indicated an ambition to play an active role as a cloud provider, although CEO Mark Zuckerberg emphasised at a shareholder meeting earlier this year that the company is still seriously evaluating the option.

The plans centre on Meta Compute, an internal organisation established in January to oversee the construction and operation of Meta's AI infrastructure. The unit is led by Santosh Janardhan, Meta's head of infrastructure, joined by Daniel Gross of the Meta Superintelligence Labs and Meta president Dina Powell McCormick.

A straightforward calculation underpins the move: Meta is investing enormous sums in data centre space and GPU capacity for its own AI applications, but will at certain points have more capacity than it consumes itself. Leasing that space to third parties is a logical step, though it also introduces new operational and commercial challenges.

Investments in the US and India run into the hundreds of billions

Meta is expected to allocate 600 billion dollars over the next three years, through to 2028, for the expansion of its American AI and data centre infrastructure. For 2026 alone, the company anticipates capital expenditure of 125 to 145 billion dollars.

Concrete projects are already under way. In Louisiana, Meta is financing a hyperscale campus of 2,250 hectares, with investor Blue Owl Capital committing 27 billion dollars. In Texas, a separate investment of 1.5 billion dollars is planned. Both sites are primarily intended for AI workloads, but also form the capacity base that Meta may potentially offer to customers.

Outside the US, Meta signed an agreement on 9 June with India's Reliance Industries for a data centre in Jamnagar, in the state of Gujarat. That facility will have a capacity of 168 megawatts, to be leased by Meta. For its energy supply, the company is working with CleanMax and Fourth Partner Energy, together accounting for nearly 1 gigawatt of renewable energy. Meta has previously invested in Reliance: in 2020 it put 5.7 billion dollars into Jio Platforms.

For comparison: the four largest hyperscalers combined, Alphabet, Microsoft, Amazon and Meta, plan to invest up to 725 billion dollars in AI infrastructure this year, according to available estimates.

Two service tiers under consideration

Meta is considering two ways to offer capacity. The first is access to its own AI models via APIs, comparable to how AWS Bedrock works: customers call models without having to manage GPU clusters themselves. The second option is leasing raw computing power, GPU-accelerated servers that customers configure themselves, the model already used by neoclouds such as CoreWeave, Nebius Group, Lambda, Crusoe and Vultr.

Anthropic and OpenAI are mentioned as potential customers, though these remain speculative at this stage; no confirmation has been provided. Meta has previously made large purchases from CoreWeave, Google and Oracle, and is therefore familiar with the market from the buyer's side. At the same time, the example of SpaceX shows that this path is viable: Elon Musk's space company already leases computing capacity from its AI infrastructure to parties including Anthropic and Google.

Uncertainty about actual demand

Whether all that capacity will truly be needed remains to be seen. Meta is building primarily for its own use: the enormous investments are intended to run Meta's own AI models and products. The question of whether there will be structurally sufficient residual capacity to sustain a fully-fledged cloud business has not yet been answered.

Entering the cloud market also requires more than available servers. AWS, Azure and Google Cloud possess extensive ecosystems of management tools, certifications, support and sales organisations that enterprise customers expect. Meta would need to build or acquire all of that, while simultaneously keeping its own AI product development on track.

The existing relationships with CoreWeave, Google and Oracle do give Meta a solid understanding of what enterprise cloud infrastructure customers require. How quickly the company moves from exploration to actual service delivery, and on what terms, is not known at this time.

What this means for the European cloud landscape

For European and Dutch parties that depend on cloud services or computing capacity for AI applications, a serious entry by Meta could open up the market further. More providers at the higher end of the market, alongside the existing hyperscalers and neoclouds, potentially means greater choice and downward pressure on prices. At the same time, European customers are becoming increasingly critical about the origin and jurisdiction of their cloud infrastructure, an area where Meta has so far announced little European data capacity. For investors in European AI infrastructure companies such as Nebius, which is already listed on the US stock exchange, Meta's entry is a signal that competition in this segment is intensifying further.