Meta's $145B AI Bet: Inside the Plan to Sell Excess Compute and Challenge AWS, Azure, and Google Cloud

Meta Platforms is making its most significant strategic pivot in years: building a cloud infrastructure business to sell access to excess AI computing power and AI models . The initiative, first reported by Bloomberg on July 1, 2026, sent Meta's stock surging roughly 10% and laid out a clear answer to Wall Street's biggest question about the company's record-breaking spending .

This is not Meta trying to replicate Amazon Web Services, Microsoft Azure, or Google Cloud. Instead, the plan is to create a specialized AI compute service—more akin to CoreWeave or SpaceX's compute offerings—aimed at companies that need GPU clusters and model hosting . Here is what we know about the plan, who is leading it, the sheer scale of the investment, and what it means for the cloud market.

The Plan: Selling Excess AI Compute

Meta Platforms is developing plans for a cloud infrastructure business that will sell access to AI computing power and models . The initiative sits under an internal umbrella called Meta Compute, which was announced in January 2026 to build and manage the company's AI infrastructure . The plans are still in development and the strategy is not yet finalized .

CEO Mark Zuckerberg first signaled the possibility at Meta's annual shareholder meeting on May 27, 2026, saying a cloud computing business was "definitely on the table" if the company's massive data center buildout left surplus capacity . On July 1, 2026, Bloomberg reported that concrete plans are now underway, sending Meta's stock up roughly 10% .

The cloud service would let developers access AI models hosted on Meta's infrastructure—including its Muse Spark model—and pay for the computing power needed to run them . Meta is reportedly evaluating two models: selling access to "raw" compute capacity (like CoreWeave) or offering AI model access similar to Amazon Bedrock .

Who Is Leading the Initiative

Meta Compute—and the cloud business line within it—is led by three senior executives :

• Santosh Janardhan — Meta's head of global infrastructure. Janardhan, who has been with the company since 2009, leads work on technical architecture, software stack, silicon program, developer productivity, and the global data center fleet and network .
• Daniel Gross — a leader inside Meta's Superintelligence Labs AI unit, reportedly leading the "superintelligence" push .
• Dina Powell McCormick — Meta's president and vice chairman, appointed in January 2026 to oversee partnerships with governments, sovereign wealth funds, and strategic investors for the company's global data center network .

Zuckerberg himself is ultimately driving the strategic direction, with CFO Susan Li managing the financial guidance .

The Scale of Data Center and AI Infrastructure Investment

Meta's capital expenditure for 2026 has been raised twice:

Guidance Date	2026 Capital Expenditure Range	Source
January 2026 (initial)	$115 billion – $135 billion
April 29, 2026 (revised)	$125 billion – $145 billion

The $10 billion increase was attributed to higher component pricing (especially memory) and additional data center costs . For context, Meta's 2025 capital expenditure was roughly $72.2 billion, meaning 2026 spending is nearly double the prior year . Meta CFO Susan Li said on the Q1 2026 earnings call that the company "continued to underestimate its compute needs" .

Beyond direct spending, Meta has signed massive multi-year infrastructure deals. In just three months (September to October 2025), Meta announced approximately $75.5 billion in infrastructure commitments :

• CoreWeave: $14.2 billion (6+ years)
• Oracle: ~$20 billion (multi-year)
• Scale AI: $14.3 billion (49% stake)
• Blue Owl/Hyperion: $27 billion (joint venture)

These deals signal that Meta is both building its own capacity and securing flexible cloud capacity for inference workloads, distributed training, and backup . The company has also committed to spending approximately $600 billion on US data centers by 2028 .

Competitive Position vs. AWS, Azure, and Google Cloud

Direct comparison is premature, but the strategic direction is clear. Here is how Meta's planned cloud business stacks up against the hyperscalers:

More like CoreWeave than AWS. Multiple outlets emphasize that Meta Compute looks less like a full-service cloud provider (IaaS/PaaS/SaaS) and more like a specialized AI compute rental service, similar to CoreWeave or SpaceX's compute offerings, aimed at AI companies that need GPU clusters and model access .

Late entrant to a mature market. AWS, Azure, and Google Cloud each have decades of enterprise cloud ecosystem, SLAs, global regions, and managed services that Meta does not yet offer. Meta would not be competing on general cloud but on AI-specific compute at scale .

Unique advantage: built-in demand. Meta already operates one of the world's largest AI fleets for its own products (Facebook, Instagram, Llama models). If it overbuilds, it can offer spare capacity at competitive prices without needing to amortize standalone sales costs .

Wall Street reaction. Investors cheered the plan as a way to monetize the $125 billion+ CapEx overhang that had been weighing on Meta's stock . The stock surged roughly 10% on the news, suggesting the market sees this as a credible ROI pathway .

Bottom line: Meta is not trying to replicate AWS's full cloud stack. It is building a focused AI infrastructure-as-a-service play to absorb surplus compute from its record-breaking data center expansion. It will compete directly with the big three for AI workloads, but only in the narrow (albeit fast-growing) segment of AI compute and model hosting rather than general-purpose cloud services .