Meta Platforms is building a cloud computing business to sell its excess artificial intelligence capacity to outside companies, according to a Bloomberg News report published July 1, 2026. The plan would pit Meta directly against Amazon Web Services, Microsoft Azure and Google Cloud, the three hyperscalers that have controlled enterprise cloud spending for two decades. Meta shares jumped as much as Meta has been investing heavily in AI infrastructure, including data centers and related technology.”
The move marks a sharp reversal for a company that spent a decade insisting its data centers existed solely to run Facebook, Instagram and WhatsApp. Meta closed out fiscal 2025 with $200.97 billion in full-year revenue, according to its FY2025 report released in December 2025, and now, after committing as much as $145 billion to AI infrastructure this year alone, Meta is looking for a way to make that spending pay for itself. Here is what the reporting shows, what Wall Street thinks, and what a fourth hyperscaler would mean for the cloud market.
Don't miss new tech stories on Google
Add Tech Insider once in the Google app and our stories appear in your news suggestions.
Meta Confirms It Is Building a Cloud Computing Business
Bloomberg broke the story on July 1, 2026, reporting that Meta was developing plans for a cloud infrastructure business, since referred to as Meta Compute, that would sell access to both AI computing power and Meta’s own AI models. Reuters, CNBC, Axios, TechCrunch and Forbes each confirmed and expanded on the report within hours, and AI Funding Tracker has since cited the Meta Compute name in its own tracking of the unit. The plans remain in development and could still change before any public launch, according to the initial reporting.
The strategic logic is straightforward. Meta has spent years building data centers sized for a future in which its own AI ambitions required nearly unlimited compute. If that buildout runs ahead of internal demand, even briefly, unused GPU capacity turns into a liability instead of an asset. Selling it converts a cost center into a revenue line, the same trade Amazon made with AWS two decades earlier.
Meta has since moved from debate to shipped product. On July 1, 2026, it launched the Meta Business Agent Platform, priced at $2 per million tokens, though usage remains free until August 1, 2026, when metered billing takes effect, according to FAQ.com.tw. Later that month, Meta also opened a public preview of its Meta Model API, offering $20 in free credits to US-based developers, per ThursdAI’s July 31, 2026 release recap. Together, the two launches confirm what earlier reporting had described only as a debate: Meta is pursuing both hosted AI model access and raw compute access at once, rather than picking one path over the other.
What a Meta Cloud Business Would Actually Sell
CNBC reported that Meta is weighing two distinct products: raw GPU compute rented directly, similar to what CoreWeave offers today, or hosted access to Meta’s own AI models through an API, closer to how OpenAI or Anthropic sell intelligence. That reporting has since played out in practice, with the July 1, 2026 launch of the Meta Business Agent Platform followed weeks later by a public preview of the Meta Model API carrying $20 in free credits for US developers, confirming the two options are not mutually exclusive and that Meta is already offering both.
“I think it would be a natural extension of our business to provide cloud services.”
Mark Zuckerberg, CEO, Meta (source)
Zuckerberg has framed the tradeoff in economic terms rather than technical ones. Selling finished intelligence carries a better margin than renting raw chips, by his own account, but raw compute is the easier product to ship first because it requires no new customer-facing model infrastructure. A bare-metal compute offering could plausibly launch well before a polished, enterprise-grade API service with the support contracts and uptime guarantees that AWS and Azure customers already expect.
Outside interest already exists. Companies have been approaching Meta directly, according to Zuckerberg, asking either to buy compute at a premium over what Meta paid for it, or to stand up API access and run workloads on Meta’s infrastructure. That inbound demand, more than any internal strategy memo, appears to be what pushed the idea from a hypothetical into a real business unit.
Zuckerberg on the Record: From “On the Table” to a Real Plan
Zuckerberg has been unusually candid about Meta’s cloud ambitions across several public appearances this year. At Meta’s annual shareholder meeting, he said a cloud computing business was:
“Definitely on the table.”
Mark Zuckerberg, CEO, Meta (source)
He also described the outside demand driving the decision, telling investors that the requests were not occasional:
“Almost every week there are different companies that come to us from outside asking us to both stand up an API service or asking if we have compute that they could buy from us at some premium to what we’ve bought it at.”
Mark Zuckerberg, CEO, Meta (source)
By July, in comments tied to Bloomberg’s reporting, he went further and explained the economics directly:
“The offers that you get for using the compute are so high that it may make sense, in some cases, to rent out or consider those kind of deals instead of your own internal uses.”
Mark Zuckerberg, CEO, Meta (source)
Even so, Zuckerberg has been clear about where he sees the bigger long-term prize. Discussing strategy around Meta’s July 29 earnings report, he said:
“We believe that there will continue to be a significantly higher margin on selling intelligence rather than selling compute directly, but we think that there’s a big opportunity obviously to sell compute as well.”
Mark Zuckerberg, CEO, Meta (source)
Taken together, the statements trace a shift from a defensive hedge to an active business plan in roughly ten weeks.
Why Now: Meta’s $145 Billion Capex Problem
The timing is not a coincidence. Meta spent $31.1 billion on capital expenditures in the second quarter of 2026 alone, according to its own earnings release. On July 29, the company narrowed its full-year 2026 capex guidance to a range of $130 billion to $145 billion, up from the $125 billion to $145 billion range it had set in April, as it kept building out its data center footprint. Cryptonomist reported on July 23, 2026 that the full $145 billion commitment is explicitly earmarked to support cloud APIs and subscription products, not just internal model training, underscoring that the buildout is meant to become customer-facing infrastructure rather than sunk cost.
That spending is consuming nearly all of Meta’s cash. Free cash flow fell about 91% year over year in the quarter, to just $784 million, as the $31.1 billion in capex ate roughly 98% of the quarter’s operating cash flow. Total costs and expenses rose 55% to $42.03 billion, partly due to a $2.4 billion charge tied to legal proceedings, which pushed full-year expense guidance up to a range of $165 billion to $169 billion.
Revenue still grew a healthy 28% year over year in the quarter. But earnings per share of $6.18 missed Wall Street’s consensus estimate of roughly $7.14 by a wide margin, according to multiple earnings trackers. A cloud business that turns idle GPUs into a revenue stream gives Meta a story to tell increasingly nervous investors about the payback period on all that spending.
The Anthropic Deal and Other Early Signals
Meta’s cloud ambitions are not purely theoretical. Reuters reported that Meta had been in talks to lease computing power to Anthropic in a deal worth up to $10 billion over two years, part of the same wave of reporting around Meta’s broader AI buildout. If finalized, it would rank among the earliest and largest examples of Meta acting as a compute supplier to another major AI lab, rather than purely as a customer or a rival.
That kind of deal also hints at how a Meta cloud business might actually launch: not as a retail product available to any developer with a credit card, but as a small number of large, negotiated contracts with AI labs and big enterprises that already need more GPU capacity than the market can supply. It is a far less risky way to test demand than building a full self-serve cloud console to rival AWS on day one.
Meta vs AWS, Azure and Google Cloud: The Numbers
Any Meta cloud business would enter a market growing faster than at any point since the pandemic-era cloud boom. All three hyperscalers posted accelerating growth in their most recent quarters, each ending around June 30, 2026.
AWS generated $42.2 billion in revenue in the second quarter, up 36.7% year over year, its fastest growth rate in 18 quarters, with operating income rising 64% to $16.6 billion. Microsoft’s Azure and other cloud services revenue grew 43% in its fiscal fourth quarter, helping push Microsoft Cloud revenue to $59.3 billion for the quarter and Azure past $100 billion in annual revenue for the first time. Google Cloud revenue reached $24.8 billion for the quarter ended June 2026, up 82% year over year, according to Reuters.
| Company | Cloud Unit | Latest Quarterly Revenue | YoY Growth | Key Detail |
|---|---|---|---|---|
| Amazon | AWS | $42.2 billion | +36.7% | Fastest growth in 18 quarters, $169B annualized run rate |
| Microsoft | Azure / Microsoft Cloud | $59.3 billion (Microsoft Cloud) | +43% (Azure) | Azure topped $100B in annual revenue for FY26 |
| Alphabet | Google Cloud | $24.8 billion | +82% | Fastest-growing of the three hyperscalers |
| CoreWeave | Neocloud (GPU rental) | $2.08 billion (Q1 2026) | +112% | $99.4B revenue backlog as of March 31, 2026 |
| Meta | Reported plan, not yet launched | Not yet reported | N/A | $31.1B quarterly capex, $130B-$145B FY26 guidance |
Meta does not report a cloud revenue line yet, because it has none. What it does report is capacity: $31.1 billion in quarterly capex and a data center buildout large enough that Zuckerberg is now openly discussing selling the leftovers.
The Neocloud Factor: CoreWeave and the Compute Resellers
Meta would not be entering an empty field. A category of GPU-focused “neocloud” providers, led by CoreWeave, has already built a business model around renting Nvidia chips by the hour to AI labs that would rather not negotiate directly with the big three hyperscalers.
CoreWeave reported first-quarter 2026 revenue of $2.078 billion, more than double the $982 million it recorded a year earlier, with a revenue backlog of $99.4 billion as of March 31, up from $25.9 billion a year before. The company has guided to full-year 2026 revenue of $12 billion to $13 billion.
Forbes has described Meta’s potential entry as a threat to exactly this segment, noting that Meta’s raw compute would compete more directly with CoreWeave and fellow neocloud Nebius than with AWS’s broader managed-services business. A hyperscaler with Meta’s balance sheet undercutting neocloud pricing on raw GPU rental could squeeze margins across the entire category, especially while GPU capacity pricing remains elevated industry-wide.
Wall Street’s Whiplash: Stock Pops, Then Drops
Meta’s stock had two very different reactions to AI-compute news within the same month. When Bloomberg’s report broke on July 1, investors treated it as good news. Axios reported shares soared nearly 9%, Livemint put the intraday gain at nearly 12% with shares touching $628 on the Nasdaq, and Yahoo Finance’s market recap logged a 7.56% close. CNBC described the reaction bluntly: the cloud report eased the biggest overhang on the stock, namely fears that AI capex would never generate a return.
Four weeks later, the mood reversed. After Meta’s July 29 earnings report showed the EPS miss, the free-cash-flow collapse and the still-massive capex range, Business Insider reported the stock tumbled 9.5%. The same capex that had investors excited about a future cloud business, in other words, spooked them once they saw the actual bill attached to it.
The whiplash captures the core tension in Meta’s pitch. A cloud business is a promising answer to “why so much capex,” but for now it remains a promise rather than a reported revenue line.
History Repeats: How AWS Began as Leftover Capacity
Meta’s pitch echoes, almost exactly, the origin story of the company it would be competing against. Amazon Web Services launched publicly in 2006, starting with Amazon S3 and Amazon EC2, after Amazon’s own engineers found that the infrastructure built to handle holiday-season e-commerce traffic sat mostly idle the rest of the year. Renting that spare capacity to outside developers eventually grew into a business that generated $42.2 billion in a single quarter two decades later.
Meta’s situation is not identical. Amazon built AWS gradually, over years, without a single stock-moving announcement. Meta is doing the reverse, announcing the strategic shift first while the actual product remains in development, according to the original Bloomberg report. But the underlying logic, that idle infrastructure is wasted money and outside customers will pay well for reliable capacity, is the same bet Amazon made two decades ago. If Meta’s cloud unit reaches even a fraction of AWS’s scale, it would validate a pattern several tech giants have now tried to repeat.
Competitive Positioning Compared
| Provider | Cloud Launched | Core Offering | Primary Customer | Reported Scale |
|---|---|---|---|---|
| AWS | 2006 | Full-stack cloud: compute, storage, 200+ managed services | Enterprises, startups, governments | $169B annualized run rate |
| Microsoft Azure | 2010 | Full-stack cloud plus enterprise software integration | Large enterprises, Microsoft 365 customers | $100B+ annual revenue (FY26) |
| Google Cloud | 2008 / 2013 | Full-stack cloud plus AI and data tooling | Enterprises, AI-native startups | $24.8B quarterly revenue, +82% YoY |
| CoreWeave | 2017 (as GPU cloud) | Raw GPU compute for AI workloads | AI labs, model developers | $99.4B revenue backlog |
| Meta (reported plan) | Not yet launched | Raw AI compute, possible model API access | AI labs, large enterprises (early deals) | $31.1B quarterly AI capex |
The comparison shows why analysts frame Meta’s likely entry point as narrower than a full hyperscaler launch. AWS, Azure and Google Cloud each sell hundreds of managed services, from databases to compliance tooling, built up over 15-plus years. Meta, at least initially, would be selling two things: chips, and potentially, model access. That looks closer to CoreWeave’s model than Amazon’s, which is why Forbes and other outlets have framed the neoclouds, not the hyperscalers, as the segment facing the most direct new competition.
What It Means for Enterprise AI Buyers
For companies burning through GPU budgets, a fourth major compute seller is good news regardless of how it gets packaged. AI labs have spent much of 2026 complaining about GPU scarcity and rising prices, a dynamic that gave existing providers pricing power. A hyperscaler-sized new entrant selling excess capacity, even in large negotiated blocks rather than through a public self-serve console, adds real supply to a market that has been supply-constrained for most of the year.
The tradeoff is dependency risk. Any enterprise leasing compute directly from Meta would be relying on a company whose core business is social media and advertising, not infrastructure uptime. Meta has never operated a public-facing cloud SLA, published a compliance certification roadmap, or run a 24/7 enterprise support desk at hyperscaler scale, the kind of groundwork behind deals like Pinterest’s multi-year AWS commitment. Those capabilities take years to build, which likely explains why the earliest reported deals, including the rumored Anthropic arrangement, involve sophisticated AI labs rather than risk-averse enterprise IT departments.
The Regulatory Questions Ahead
A social media giant with billions of daily users turning its data center footprint into a cloud computing business is likely to draw scrutiny beyond the usual competitive response from AWS and Azure. Meta already operates under active antitrust oversight in the United States and ongoing obligations under the European Union’s Digital Markets Act. Regulators who have spent years scrutinizing how Meta uses its scale in social networking may end up asking similar questions about how it uses its scale in AI infrastructure, particularly if Meta bundles compute access with preferential terms for its own AI models.
None of that is likely to slow Meta down in the near term. The plans reported so far involve negotiated enterprise deals, not a mass-market product that would immediately trigger the kind of consumer-facing scrutiny Meta’s other businesses attract. But a cloud unit growing alongside existing regulatory pressure is a detail investors and regulators alike are likely to keep watching closely.
5 Predictions for Meta’s Cloud Business
- Compute first, console later. Expect raw GPU rental deals, similar to the reported Anthropic arrangement, well before any self-serve API console resembling AWS or Azure appears.
- AI labs sign first, enterprise IT signs later. Sophisticated buyers who need capacity now and can tolerate less mature support terms will move first, easing Meta into the business without requiring hyperscaler-grade compliance infrastructure on day one.
- Neoclouds feel the pressure before AWS does. Meta’s likely raw-compute model competes most directly with CoreWeave and Nebius, not with the managed-services businesses that make up most of AWS and Azure’s revenue.
- The unit needs to show quarterly proof fast. With free cash flow down 91% year over year in Q2, investors will want to see real cloud revenue within two or three quarters, or risk treating the idea as just another capex justification story.
- Others follow if it works. Companies sitting on large AI infrastructure investments, from Oracle to well-funded AI labs building their own data centers, are likely to explore similar “sell the excess” plays if Meta’s experiment shows real demand.
Related Coverage
- Data Center Leases Hit $850B, Meta, Microsoft Lead [2026]
- Qualcomm’s Dragonfly C1000 Lands Meta, Eyes $15B [2026]
- Microsoft Azure Earnings: $678B Backlog, Up 84% [2026]
- Google Cloud Hits 82% Growth Ahead of AWS Earnings [2026]
- Cloud Waste Hits 29% as AI Spend Breaks Budgets [2026]
- Pinterest Commits $4B to AWS Through 2031 [2026]
- AWS Hikes EC2 GPU Pricing 20%, Second Time in 2026
Frequently Asked Questions
Is Meta actually launching a cloud computing service?
Meta has not announced a public launch. Bloomberg reported on July 1, 2026 that Meta is developing plans for a cloud business, and Zuckerberg has confirmed the idea is under active consideration, but the company has not published pricing, a product name or a launch date. The plans are described as still in development.
When did the Meta cloud business news break?
Bloomberg News first reported the plans on July 1, 2026. Reuters, CNBC, Axios, TechCrunch and Forbes published their own coverage the same day, and Zuckerberg addressed the topic again around Meta’s July 29 earnings report.
Why did Meta stock jump on the cloud business report?
Investors read the news as a sign that Meta’s massive AI capex could eventually generate direct revenue rather than only supporting internal products. Reported gains on July 1 ranged from 7.56% to nearly 12%, depending on the source and the moment of measurement.
How much is Meta spending on AI infrastructure in 2026?
Meta spent $31.1 billion on capital expenditures in the second quarter of 2026 and guided to $130 billion to $145 billion for the full year, narrowed from an earlier $125 billion to $145 billion range.
Will Meta compete with AWS and Azure, or with CoreWeave?
Most likely both, but not equally. Analyst commentary suggests Meta’s initial raw-compute offering would compete most directly with GPU-focused neoclouds like CoreWeave and Nebius, while any future model-hosting API would compete more directly with AWS, Azure and Google Cloud’s AI services.
What did Mark Zuckerberg say about the cloud plans?
Zuckerberg has called a Meta cloud business “definitely on the table,” described frequent inbound requests from outside companies wanting to buy compute or API access, and said in a Bloomberg-linked interview that current compute offers are high enough that renting out capacity “may make sense” in some cases.
Is the Anthropic-Meta compute deal confirmed?
Not officially. Reuters reported that Meta had been in talks to lease computing power to Anthropic in a deal that could be worth up to $10 billion over two years, but neither company has published contract terms.
What happened to Meta’s stock after its Q2 2026 earnings?
Meta shares fell about 9.5% after the July 29 earnings report, as investors weighed an EPS miss, a 91% year-over-year drop in free cash flow and continued heavy capex guidance against the company’s 28% revenue growth.
