TL;DR — Meta Platforms is building a cloud computing business called Meta Compute, planning to sell AI compute and hosted models to outside customers. Backed by $115-135B in annual AI capex and a GPU fleet few can match, Meta is taking on AWS, Azure, and Google Cloud directly. Shares jumped 9.3% on the July 1 Bloomberg report. The initiative is led by infrastructure head Santosh Janardhan, superintelligence lead Daniel Gross, and Meta President Dina Powell McCormick.
Introduction
Meta Platforms is making its boldest infrastructure play yet: a cloud computing business that will sell AI computing power and hosted models to outside customers, directly challenging Amazon Web Services, Microsoft Azure, and Google Cloud.
The initiative, known internally as Meta Compute, was first reported by Bloomberg on July 1 and sent Meta shares up 9.3% — their biggest intraday gain since April. The ripple hit competitors fast: neocloud provider CoreWeave fell as much as 14%, and Nebius Group dropped 17%.
What Meta Compute Will Offer
According to people familiar with the plans, Meta is pursuing a two-pronged strategy:
-
Model-as-a-Service: Selling API access to AI models hosted on Meta’s infrastructure — including its proprietary Muse Spark family — similar to AWS Bedrock or Azure AI Foundry. Developers would pay per-token to use Meta’s models without managing any hardware.
-
Raw Compute Rental: Selling bare-metal access to GPU capacity, akin to what neocloud providers like CoreWeave offer. This would let customers train or run their own models on Meta’s data center fleet.
The initiative is led by a trio with serious operational weight: Santosh Janardhan (Meta’s head of infrastructure), Daniel Gross (a leader within Meta Superintelligence Labs), and Dina Powell McCormick (Meta’s President).
The $115 Billion Rationale
Meta has committed between $115 and $135 billion in capital expenditure for 2026 alone, overwhelmingly directed at AI infrastructure — data centers, GPUs, and networking. The company has been stockpiling compute at a scale that has made Wall Street nervous about returns.
A cloud business gives that spending a monetization path. As Mark Zuckerberg told shareholders in May:
“It’s definitely on the table. Almost every week there are different companies that come to us from the outside asking us to both stand up an API service or asking if they have compute that they could buy from us at some premium to what we’ve bought it at.”
The demand is real. Major AI developers face persistent compute shortages, and the neocloud market — led by CoreWeave, which went public in 2025 — has validated the model of renting GPU capacity to AI companies. SpaceX (via xAI) already demonstrated the playbook, renting its Memphis data center to Anthropic and striking a deal with Google, on track toward a projected $50 billion in cloud revenue by 2028.
(Sources: Bloomberg — Meta Is Building a Cloud Business, LA Times — Meta shares surge, CNBC — Meta cloud plans)
From Open-Source Champion to Cloud Provider
The cloud push completes a dramatic strategic pivot for Meta’s AI division. Three years ago, Meta was the loudest advocate for open-weight AI, distributing Llama models freely to millions of developers. In April 2026, it abandoned that approach entirely, launching Muse Spark as a proprietary, cloud-only model with no downloadable weights.
Now Meta is positioning to become the store as well as the product — not just building AI models, but renting the infrastructure and selling API access to those models. It’s the same vertical integration play that made AWS dominant: own the hardware, own the models, own the customer relationship.
The developer community Meta spent three years cultivating with open-weight Llama releases may not celebrate this one. Having already lost access to downloadable models, developers now face the prospect of being charged per-token to use Meta’s AI through a proprietary cloud — the exact model Meta once positioned itself against.
The Competitive Landscape
Meta enters a market dominated by three incumbents that spent decades building cloud platforms:
| Provider | Market Share | Key Strengths |
|---|---|---|
| AWS | 31% | Market leader with Bedrock, SageMaker, broadest enterprise base |
| Microsoft Azure | 24% | Deeply integrated with OpenAI models and enterprise accounts |
| Google Cloud | 12% | Leading AI/ML tooling and TPU infrastructure |
Meta’s advantage is its infrastructure scale — the company has been buying GPUs at a rate few can match — and its ownership of frontier models through Muse Spark. The disadvantage is that building enterprise sales teams, support operations, and developer platforms is a fundamentally different business from running a social media company.
Wall Street’s initial reaction suggests optimism. CNBC reported that analysts see the cloud business as “a credible path to monetizing Meta’s massive AI investments,” even if margins will be lower than Meta’s advertising business.
(Source: TechCrunch — Meta cloud business plans)
What’s Next
The plans are still in development and could change. Meta declined to comment officially. But the strategic logic is hard to miss: having spent more on AI infrastructure than almost any company on earth, Meta is looking to sell access to anyone willing to pay — and reshape the cloud computing market in the process.
FAQ
Q: Is Meta Compute already available?
No. The plans are still in development and could change. Meta has not announced a launch timeline or pricing.
Q: What models will Meta Compute offer?
The centerpiece is Muse Spark, Meta’s proprietary model family launched in April 2026. The service will also likely host other models, following the Bedrock/AI Foundry model of multi-model API access.
Q: Why is this happening now?
Meta has committed $115-135B in capex for 2026, mostly on AI infrastructure. A cloud business gives Wall Street a monetization narrative for that spending, and Zuckerberg confirmed in May that external companies were already asking to buy compute or API access.
Q: Does this mean Meta abandoned open-source AI entirely?
Yes, at least for now. Meta stopped releasing open-weight models in April 2026 when it launched Muse Spark as a proprietary, cloud-only model. Meta Compute extends that strategy — Meta wants to be the store, not just the factory.
Q: How does this compare to AWS, Azure, and Google Cloud?
Meta has comparable GPU scale but none of the enterprise sales infrastructure. The cloud incumbents have spent 15+ years building developer platforms, support operations, and customer relationships. Meta starts from zero on the enterprise side.
Further Reading
- Bloomberg — Meta Is Building a Cloud Business to Sell Excess AI Compute (July 1, 2026)
- CNBC — Meta shares jump on report of plans to build AI cloud business (July 1, 2026)
- LA Times — Meta explores cloud computing business, shares surge (July 1, 2026)
- TechCrunch — Meta plans cloud business to compete with AWS, Microsoft (July 1, 2026)
- /2026/06/meta-ai-strategy-2026-muse-spark-open-source-pivot/ — TAR’s previous coverage of Meta’s AI pivot