Facing investor skepticism over its artificial intelligence spending, Meta is taking a page from CoreWeave’s playbook.

Since the advent of the AI boom, Mark Zuckerberg’s firm has kept pace with hyperscale cloud providers Amazon, Microsoft and Google in the data center spending race. Meta has invested hundreds of billions of dollars on AI build-out, delivering two of the first gigawatt-scale campuses in Ohio and Louisiana.

Unlike the cloud giants to whom it is so often compared, Meta hasn't been building AI infrastructure to meet demand from external cloud customers.

Instead, the firm’s investment in gigawatts of computing capacity has almost exclusively been intended to power its own AI ambitions: improving the advertising and recommendation engines behind Facebook, Instagram and WhatsApp as well as developing consumer-facing products and services.

Yet in recent weeks, Meta has made a sudden pivot toward becoming an AI cloud provider.

Zuckerberg outlined plans on Meta's late July earnings call for it to begin “selling compute directly” to other large AI firms — effectively announcing the company’s entry into the fast-growing landscape of AI-focused cloud firms known as neoclouds.

While cloud offerings will account for only a portion of Meta’s computing power, the decision to sell capacity to external parties at all marks a dramatic course change for a company that had shown little interest in the idea in the past. It’s a shift that Zuckerberg, speaking with Wall Street analysts last month, insisted was born out of market opportunity more than necessity.

“There's just nowhere near enough compute for all the demand,” Zuckerberg said. “We are getting a large number of offers for the compute that we have.”

However, industry analysts say Meta’s decision to launch cloud offerings is a response to growing Wall Street anxieties about a lack of return on Big Tech’s AI investments. After tech companies reported second-quarter earnings, investors punished firms — Meta among them — that increased AI capex without demonstrating commensurate returns or a clear path to profitability.

While Meta isn't the only firm whose spending is outpacing its AI revenue growth, the company’s core business model puts it at a disadvantage relative to its Big Tech brethren when it comes to short-term revenue. Meta, unlike the cloud providers, can’t show a sales backlog demonstrating that, even if it is still operating at a loss, each new kilowatt of computing capacity will directly translate into new revenue.

"Meta is spending like a hyperscaler without a hyperscaler's business model," Josh Gilbert, eToro's lead APAC analyst, told Reuters. "Microsoft, Alphabet and Amazon can point their data centre dollars at cloud businesses that sell compute straight back out the door, but Meta doesn't ​have the same outlet, so every dollar of ​build-out leans on the ads business."

Becoming an AI cloud provider, analysts say, gives Meta a quantifiable near-term revenue flow that may be necessary to keep Wall Street from jumping ship as the firm's spending continues to climb, with an anticipated $145B in capex this year alone.

“I think that this is a response to complaints that the company may be overspending and skepticism that Meta will ever earn a commensurate return on its capex,” said Paul Meeks, head of technology research at Freedom Capital Markets, according to CNBC. “The problem with this company is that it only builds, or only thus far, capacity for itself, and it’s not really monetizing any AI apps yet.”

Meta has yet to provide any specifics as to timeline, potential product branding or deals that may be in the pipeline. But the firm’s leadership did outline a two-pronged strategy for quickly monetizing some of Meta’s vast AI infrastructure investments.

In part, Meta will begin offering cloud access to its AI models, a business line that would be a competitor to the major “AI labs” like OpenAI and Anthropic. At the same time, Meta is looking to sell its computing capacity to other major large-scale AI firms, a model that would instantly make Meta a serious competitor to CoreWeave and other large neoclouds.

“I don’t think this is them going after the cloud businesses — you’d need a lot to go after AWS or Azure,” Evercore ISI’s Mark Mahaney said in an interview with CNBC. “But this is going after the neocloud businesses.”

Neoclouds represent the fastest-growing segment in the booming data center market. With major players including CoreWeave, Core Scientific, Nscale and Lambda, these AI-specific cloud providers offer on-demand access to the graphics processing unit computing critical for AI. The sector posted 223% year-over-year revenue growth in the fourth quarter, according to Synergy Research Group.

The neocloud ecosystem has more than 190 operators, most of them startups, according to JLL. And while these companies have a number of different business models — catering to different-size companies with contracts ranging from hours to years — the largest players in the space, like CoreWeave, have anchored their businesses by selling large blocks of capacity to hyperscalers.

It’s this segment of the market that Meta is targeting, at least initially, analysts say. Reuters reported last month that Meta was negotiating a deal to lease computing capacity to Claude-maker Anthropic that would be worth $10B over two years. Meta hasn't confirmed those reports, but Zuckerberg did tell analysts the company has had talks with potential customers for deals that would have delivered strong return on Meta’s infrastructure spending.

“We're getting a lot of offers for compute at a significant premium over what we paid for it,” Zuckerberg said.

This kind of cash infusion would be a timely boost for Meta following a quarter in which revenues fell short of investor expectations. The firm’s free cash flow dropped 91% during the second quarter, due mainly to its massive AI spending, helping send its share price down more than 8% from the start of the year.

While Meta is becoming a competitor to the major neoclouds, it is also one of the segment’s largest customers.

In 2025, Meta entered into a $14.2B cloud contract with CoreWeave that was expanded to $21B in April. The contract runs through 2031, with a possible extension to 2032. Meta has also purchased capacity from other neoclouds, including AMD and Crusoe.

Meta is also not the first AI company to begin selling its computing capacity in recent months. SpaceX’s AI division, formerly xAI, initially built out infrastructure to support its own AI offerings but has since started selling excess capacity to external customers like Google and Anthropic. Evercore ISI’s Mahaney said these deals are generating around $20B in annual revenue for SpaceX and may have helped inspire Meta’s pivot.

But others frame these moves toward the cloud as something of a strategic retreat, as Zuckerberg and Musk’s firms pivot to selling compute as a commodity to third parties after their own consumer AI businesses failed to generate the returns investors had hoped for.

[Both Meta and SpaceX] failed to bring to market an AI model that drove huge customer traction,” said Brian Schechter, a partner at Primary Venture Partners, according to CNBC. “Being able to monetize their compute after a missed training run shows how compute can function more like a commodity.”