Meta builds 'Meta Compute' cloud to sell surplus AI capacity

The race to build the infrastructure required to power the next generation of artificial intelligence has driven unprecedented levels of capital expenditure. Technology giants are investing hundreds of billions of dollars in data centres, specialised chips and energy resources.

However, as the initial wave of infrastructure development matures, companies are exploring new avenues to generate returns on these massive investments. Meta Platforms Inc. is reportedly developing plans to launch a cloud infrastructure business, aiming to sell access to its excess AI computing power and hosted models.

This strategic pivot, mirroring a similar move by SpaceX's xAI, suggests that the data centre build-out is evolving into a rentable asset class. For the broader industry, including project delivery professionals involved in constructing these facilities, this development provides crucial insight into who will fund and build the next wave of capacity.

Meta has been exceptionally aggressive in securing the hardware necessary to support its artificial intelligence ambitions. By the end of the first quarter of 2026, the company had committed to spending $182.9 billion on AI infrastructure over the coming years. This includes massive, ongoing data centre projects in locations such as Louisiana and Ohio, with the latter expected to be the size of Manhattan upon completion.

Despite this colossal investment, Meta has not seen the same level of direct end-user demand for its AI models and services as competitors like Google or OpenAI. The company does not break out revenue specifically for its Meta AI or its open-weight Llama models, and its executives have primarily emphasised the internal, corporate applications of these technologies.

To recoup its investment, Meta is preparing to offer its raw computing capacity to outside customers. The new initiative is reportedly dubbed "Meta Compute" and includes several key components:

• Raw compute capacity: Selling direct access to computing power, similar to CoreWeave's business model

• Hosted AI models: Offering access to proprietary models, including the recently launched closed-weight model Muse Spark

• Enterprise-grade services: Providing the infrastructure and support needed for external customers

• Leadership team: Spearheaded by Santosh Janardhan (head of infrastructure), Daniel Gross (Meta Superintelligence Labs leader) and Dina Powell McCormick (president)

The entry of Meta into the cloud infrastructure market introduces a formidable new competitor to established players like AWS, Google Cloud and Microsoft Azure. While these incumbents have long dominated the general cloud computing space, the specific demand for AI-optimised compute provides an opening for new entrants with massive, purpose-built capacity.

Meta's approach is not entirely without precedent in the current market cycle. In early May 2026, SpaceX's xAI signed a deal with Anthropic to lease all the compute capacity at its Colossus 1 data centre and has since signed similar agreements with Google and Reflection AI. This trend indicates that the true winners of the artificial intelligence boom may not necessarily be the developers of the best models but the entities that own and control the underlying physical infrastructure.

The assessment highlights the strategic positioning of Meta's move within the broader competitive landscape, where infrastructure ownership is increasingly viewed as a core competitive advantage.

The development of Meta Compute has significant implications for the data centre industry and the project delivery professionals responsible for building these facilities. Consider the following factors:

• Sustained demand: Treating AI compute as a rentable commodity could stabilise demand for new construction, even if direct consumer adoption of AI tools fluctuates

• Capacity concerns: Some analysts warn that the rapid build-out could lead to a glut, particularly if hardware depreciates faster than it generates revenue

• Asset utilisation: Maximising utilisation through third-party leasing is a proven strategy in the technology sector and could improve returns on capital

• Market signals: If major players like Meta are already looking to offload excess capacity, it may signal that internal demand is not scaling as quickly as the infrastructure itself

McKinsey's analysis of the compute market provides additional context. The firm estimates that "in a constrained-demand scenario, AI-related data centre capacity could require $3.7 trillion in capital expenditures." This figure underscores the scale of the infrastructure challenge and the potential for companies like Meta to generate substantial revenue by monetising their capacity.

Meta's decision to launch a cloud business aligns with statements made by CEO Mark Zuckerberg earlier in the year, where he noted that such a move was "definitely on the table" as a way to generate returns on the company's investments in "superintelligence."

The success of Meta Compute will likely depend on the company's ability to offer competitive pricing and reliable service compared to the established cloud giants. While Meta has vast experience in managing infrastructure for its own global applications, providing enterprise-grade cloud services to external customers requires a different operational focus and support structure.

The competitive landscape is intensifying. AWS, Google Cloud and Microsoft Azure have established relationships with enterprise customers and deep expertise in cloud service delivery. Meta's advantage lies in its ability to offer purpose-built AI infrastructure at scale, but it must overcome the perception that it is a newcomer to the enterprise cloud market.

• Meta's plan to sell excess computing power indicates a strategic shift towards monetising its $182.9 billion infrastructure investment.

• The move positions Meta as a direct competitor to established cloud providers like AWS, Google Cloud and Microsoft Azure in the AI compute space.

• Treating AI data centres as rentable asset classes may sustain construction demand, even if direct consumer adoption of AI models slows.

• The trend of technology companies leasing out their purpose-built infrastructure suggests that owning the physical assets is becoming as important as developing the software.

• Meta's success in this new venture will require adapting its operational model to support enterprise-grade, external customer services.

To stay informed about the evolving economics of artificial intelligence infrastructure and its impact on the construction sector, subscribe to the Project Flux newsletter.

All content reflects our personal views and is not intended as professional advice or to represent any organisation.