Meta Platforms is building a cloud infrastructure business designed to sell access to its AI computing power and models to external customers. The initiative, internally called “Meta Compute,” puts the social media giant on a direct collision course with Amazon Web Services, Microsoft Azure, and Google Cloud.

The project was formalized internally in January 2026. CEO Mark Zuckerberg confirmed outside interest in the offering during Meta’s annual shareholder meeting on May 27, 2026, noting that external firms had expressed interest in Meta’s services, including APIs and leased computing infrastructure.

From social media company to cloud provider

Here’s the thing about Meta’s cloud ambitions: the company has been spending like a sovereign wealth fund on AI infrastructure. Meta’s projected capital expenditure for AI and data-center growth in 2026 sits between $125 billion and $145 billion. That’s not a typo. That range alone, roughly $20 billion wide, is larger than the entire annual revenue of most tech companies.

All that spending was originally meant to power Meta’s own AI products, from recommendation algorithms on Instagram and Facebook to its Llama large language models. But when you build that much computing infrastructure, you inevitably end up with excess capacity. And excess capacity is just money sitting on a shelf.

The logical next step is renting it out. Amazon figured this out two decades ago when it realized its e-commerce infrastructure could serve other businesses. AWS went from internal tool to the most profitable division at Amazon. Meta appears to be running the same playbook, just with AI compute instead of general cloud services.

Meta is also expected to introduce multi-gigawatt data-center clusters as part of this strategy, a scale of infrastructure that would rival the largest facilities operated by existing hyperscalers.

Why this matters for the AI economy

The cloud infrastructure market is dominated by three players. AWS, Azure, and Google Cloud collectively control the vast majority of enterprise cloud spending. Breaking into that oligopoly is extraordinarily difficult, and most companies that have tried have either failed or carved out niche positions.

Meta’s angle is different, though. The company isn’t trying to sell general-purpose cloud computing. It’s targeting AI workloads specifically, a segment of the market that’s growing faster than anything else in enterprise tech. Companies training and deploying AI models need GPU clusters, and there simply aren’t enough of them available.

That supply-demand imbalance gives Meta a genuine opening. If you’ve already built the infrastructure and your own AI products don’t consume all of it, selling the surplus is essentially free revenue on hardware you’ve already paid for. The margins on cloud compute are notoriously attractive, which is why AWS generates more operating profit than Amazon’s entire retail business.

Look, Meta isn’t the only company eyeing this opportunity. Oracle, CoreWeave, and a growing list of AI-focused cloud providers have been gaining traction by focusing specifically on GPU compute. But none of them have Meta’s scale of investment or its existing AI model ecosystem built around Llama.

The combination of raw compute capacity and proprietary AI models could be a compelling package for enterprise customers who want both infrastructure and ready-made AI tools.

What this means for investors

The market’s initial reaction to the news was positive, with Meta’s share price getting an immediate lift. That response makes sense. Cloud infrastructure businesses generate recurring revenue with high margins, exactly the financial profile that Wall Street rewards with premium valuations.

But there are real risks here. Building a world-class cloud business requires more than just hardware. It demands enterprise sales teams, service-level agreements, 24/7 support operations, compliance certifications, and the kind of reliability guarantees that companies like AWS have spent years perfecting. Meta has none of that institutional muscle today.

There’s also the question of whether Meta’s biggest potential customers, companies that compete with it in advertising or social media, would feel comfortable running their AI workloads on Meta’s infrastructure. The cloud computing market runs partly on trust, and Meta’s history with data privacy doesn’t exactly inspire universal confidence.

For the broader tech landscape, Meta’s entry adds another deep-pocketed competitor to the AI infrastructure race. That’s good news for companies that consume compute, since more supply should eventually moderate pricing. It’s less good news for smaller, specialized cloud providers that have been thriving precisely because demand outstrips supply from the big three.

The $125 billion to $145 billion capex figure is the number to watch. If Meta Compute gains traction with external customers, that spending transforms from a cost center into a revenue-generating platform. If it doesn’t, it’s one of the largest infrastructure bets in corporate history with limited upside beyond Meta’s own products. The difference between those two outcomes is worth hundreds of billions of dollars in market capitalization.