Meta dives into AI cloud: will rent out its computing power

What happened?

Mark Zuckerberg has publicly confirmed that Meta is exploring an AI cloud business to rent out its computing power. According to Bloomberg, Meta's CEO stated that it 'makes sense' to sell access to the company's computing capacity. The official confirmation puts a name to a plan that had already been leaked earlier this month, when it was reported that Meta was evaluating an internally named venture Meta Compute to rent out its spare AI capacity. This is not Meta's first foray into cloud services; in 2022, the company had already launched Meta Cloud, a limited offering for select partners, but the new project targets a mass market.

Why is it important?

This move positions Meta as a direct competitor in the AI cloud infrastructure market, currently dominated by AWS (with a 32% share), Microsoft Azure (23%), and Google Cloud (11%), according to Synergy Research Group data from 2025. Meta has made massive investments in specialized hardware, such as its own MTIA chips (Meta Training and Inference Accelerator) and AI-optimized servers, designed to train and run its large language models like Llama 4, which requires clusters of 16,000 GPUs. According to Meta's 2025 annual report, the company spent over $35 billion in infrastructure capex, of which 70% was allocated to AI. By renting out idle capacity, Meta not only generates a new revenue stream but also democratizes access to high-performance computing power for startups and researchers, who often face waitlists at traditional hyperscalers.

Historically, Meta has followed a strategy similar to Amazon's with AWS: build infrastructure for internal use and then commercialize the surplus. In 2013, Amazon began renting out its spare computing capacity, giving rise to AWS, which today generates over $100 billion annually. Meta, with its enormous AI investment, seems to replicate that model, albeit in a more mature market with established competitors. The key difference is that Meta not only sells computing power but also promotes its open Llama model ecosystem, which could create a virtuous cycle: more use of Llama attracts customers to Meta Compute, and vice versa.

Market consequences

Meta's entry could intensify the price war in AI cloud, benefiting consumers but pressuring margins of traditional hyperscalers. According to a Bernstein analysis, rental prices for GPUs like the NVIDIA H100 have dropped 40% in the past year due to oversupply, and Meta's arrival could accelerate this trend. However, by offering its own infrastructure, Meta could incentivize the use of its Llama model, creating a vertically integrated ecosystem similar to Apple's (hardware + software + services). This would put pressure on companies like OpenAI, which relies on Azure, and on Google, which trains its models on its own cloud but also rents out capacity.

For startups, the additional competition means more options and potentially lower prices, but also introduces risks. Meta has a history of abrupt changes in its platform policies, such as the closure of APIs in 2018 after the Cambridge Analytica scandal. Companies renting capacity from Meta Compute could face sudden changes in service terms or pricing. Moreover, doubts persist about the security and privacy of customer data when processed on servers of a company whose core business is data-driven advertising. Although Meta has established technical barriers between its divisions, trust is a difficult asset to regain. A 2025 Gartner report notes that 60% of CIOs cite privacy as the main barrier to adopting cloud services from companies with data-driven business models.

Another potential impact is on the chip supply chain. Meta, like other hyperscalers, has designed its own MTIA chips to reduce dependence on NVIDIA. If Meta Compute succeeds, it could accelerate the adoption of custom hardware, pressuring NVIDIA to innovate faster or lower prices. This would also affect other chip providers like AMD and Intel, which seek to gain share in the inference market.

What readers should know

• Meta Compute is still an exploratory project; no launch date or pricing has been defined. Sources close to the company indicate the service could launch in beta by late 2026, according to The Information.
• The company already rents capacity through partnerships, such as the one with Microsoft to run Llama on Azure, but a public service would be a significant step, making it a direct competitor to its own partners.
• Meta faces regulatory and trust challenges that could limit its enterprise adoption. The European Union has already launched an investigation into potential anticompetitive practices if Meta ties Llama usage to compute discounts.
• The AI cloud market is growing at a compound annual rate of 35% and is expected to reach $200 billion by 2028, according to MarketsandMarkets. Even a small share for Meta could generate significant revenue: a 5% market share would mean $10 billion annually.

“Selling access to computing power makes sense,” Zuckerberg said, according to Bloomberg.

In summary, Zuckerberg's confirmation marks a milestone in Meta's strategy to monetize its AI investment. While the path is fraught with technical, regulatory, and trust challenges, the AWS precedent shows that a hyperscaler can transform its internal infrastructure into a multi-billion dollar business. The key question is whether Meta can overcome privacy skepticism and whether its Llama ecosystem will be attractive enough to shift market dynamics. The coming months will be crucial to see if Meta Compute moves from rumor to a reality that redefines competition in the AI cloud.