By Maksym Misichenko · CNBC ·
What AI agents think about this news
The panelists generally agree that Meta's entry into the cloud infrastructure space is risky, with potential margin compression and regulatory challenges outweighing short-term revenue gains. They express concern about Meta's ability to differentiate, monetize excess capacity, and avoid antitrust scrutiny.
Risk: Regulatory scrutiny and potential 'self-preferencing' investigations due to Meta's internal compute allocation and prioritization of its own models.
Opportunity: Monetizing excess AI capacity to justify ongoing capex, although panelists express skepticism about the feasibility and profitability of this strategy.
This analysis is generated by the StockScreener pipeline — four leading LLMs (Claude, GPT, Gemini, Grok) receive identical prompts with built-in anti-hallucination guards. Read methodology →
Meta CEO Mark Zuckerberg appears poised to make a big bet on a potentially big market, but one that commands much slimmer margins than his company's dominant online ad business.
Cloud infrastructure has proven to be highly lucrative for hyperscaler peers Amazon, Microsoft and Google, and Zuckerberg has hinted of late that Meta could be headed in that direction. On Wednesday, CNBC's Jim Cramer confirmed that Meta will sell excess computing power to outside customers. The company is debating whether to offer access to AI models hosted on its infrastructure or to sell access to raw computing power, according to Bloomberg.
Wall Street welcomed the news. After slumping for the past year, Meta's stock started the third quarter with a bang, jumping 9% on Wednesday for its sharpest rally in more than five months. Investors have been looking for Meta to diversify its business and monetize its multi-hundred-billion-dollar investment in advanced data centers and artificial intelligence infrastructure.
"Making this as a revenue stream has been part of their road map," said Karan Ramchandani, managing director at advisory firm Post Oak Group. "It seems like a no-brainer to compete in the market, to sell compute power to other B2B players."
At Meta's annual shareholder meeting in May, Zuckerberg said a potential cloud computing business is "definitely on the table." And seven months earlier, on an earnings call, Zuckerberg said companies are regularly "asking if we have compute that they could buy from us at some premium to what we've bought it at."
Just before Wednesday's rally, Meta's stock closed out its fourth straight quarterly drop. In April, Meta boosted the high end of its 2026 capital expenditures guidance by $10 billion to $145 billion. Some of that is getting funded through debt, with the company raising $25 billion from a bond sale just as it was reporting first-quarter earnings.
"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, regarding Meta's push into cloud. "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."
Almost all the financial benefits of Meta's AI spending to date have been recognized in the company's core advertising business, which has seen dramatically improved targeting capabilities and has offered a wider suite of creative tools to marketers. Meta still gets 98% of its revenue from digital ads.
Zuckerberg has been trying to change the narrative, with cloud being perhaps the most ambitious new effort. Meta shares jumped almost 4% in May, when the company announced various paid subscription plans for Instagram, Facebook and WhatsApp in addition to two subscription services for its Meta AI app and website.
Meta declined to comment for this story.
Cloud infrastructure is a particularly valuable commodity as the generative AI boom nears its fourth anniversary, and is something that very few companies can afford to provide at scale. The U.S. leaders are Amazon Web Services, Microsoft Azure and Google Cloud, which have all built large businesses by allowing companies to offload their computing requirements.
Mark Mahaney, an analyst at Evercore, said it's unlikely that Meta will try to challenge those hyperscalers. Rather, Mahaney sees Meta following in the footsteps of so-called neoclouds such as CoreWeave and Nebius, which offer access to AI-specific computing products like Nvidia chips and systems.
Shares of CoreWeave and Nebius both suffered double-digit drops on Wednesday following the Meta report.
Mahaney said Meta may have been motivated in part by Elon Musk's SpaceX. The company, which owns xAI, has recently signed deals to offer capacity to Google and Anthropic amounting to more than $2 billion in combined revenue a month as well as to startup Reflection AI.
Brian Schechter, a partner at Primary Venture Partners, also made the comparison to SpaceX. He said the companies are similar in that they've spent billions of dollars training big AI models on top of their own infrastructure.
Both companies "failed to bring to market an AI model that drove huge customer traction," Schechter said. "Being able to monetize their compute after a missed training run shows how compute can function more like a commodity."
One area of concern for some investors will be the potential hit to Meta's profitability. Selling cloud services typically requires building a big enterprise sales and support team, and the margins can't match what Meta generates from ads.
Meta's gross margin of 82% is among the highest in the tech industry, and the company recorded an operating margin of 41% in the latest quarter. Google provides a glimpse of what's to come.
Google's services business, which mostly comes from ads, notched an operating margin of 42% in the first quarter, while for cloud the margin was 18%. It took many years just to get there. The company launched its cloud infrastructure business in 2008 and made it generally available in 2011. In 2020, Google started disclosing financials, and didn't record a profit until the first quarter of 2023.
Meeks said that while Meta "probably has one of the most glorious business models in tech," anything it enters outside of online ads "would be dilutive to their business and would lower their margins from their glory days."
"As a Meta shareholder, I'd rather see them continue with open models and monetize AI through products and services with much higher margins than get into the brutal battle of building data centers in places like North Dakota," Meeks said.
WATCH: Meta building out cloud business is 'a really smart pivot,' says Evercore ISI's Mark Mahaney.
Four leading AI models discuss this article
"Meta's pivot to cloud infrastructure is a margin-dilutive strategy that signals an inability to monetize their massive AI capital expenditures through their core advertising business alone."
Meta entering the cloud infrastructure space is a defensive maneuver disguised as a growth opportunity. While the market cheered the 9% pop, investors are conflating 'utilization' with 'profitability.' Meta’s current 41% operating margin is powered by a high-margin ad tech stack; pivoting to a commodity compute provider introduces massive OpEx friction—enterprise sales, support, and maintenance—that will inevitably compress margins. The real risk is that this is a 'distress sale' of excess capacity rather than a strategic product launch. Unless Meta can differentiate beyond being a 'neocloud' provider, they are essentially subsidizing the AI infrastructure of their competitors while diluting their own best-in-class financial profile.
If Meta successfully monetizes idle GPU clusters that would otherwise depreciate, they improve their return on invested capital (ROIC) without needing to build a traditional enterprise sales force, effectively turning a sunk cost into a high-margin cash stream.
"Meta is monetizing failure (unused AI infrastructure) as success, and Wall Street is rewarding the narrative despite the margin math being structurally unfavorable versus the core ad business."
Meta's cloud push is being priced as a growth diversification story, but the article itself contains the refutation: Google Cloud took 15 years to reach 18% operating margins versus Meta's current 41%. Meta is burning $145B capex annually with no proven AI product monetization outside ads (98% of revenue). The 9% rally reflects investor desperation for a narrative, not economics. Selling commodity compute to neoclouds and AI startups is a race-to-the-bottom margin business. Meta's real problem—that $25B debt raise signals—is that it's built infrastructure without a customer. Renting spare capacity doesn't solve that; it subsidizes competitors' training runs.
If Meta can undercut AWS/Azure on GPU pricing by 15-20% due to lower cost-of-capital and in-house chip design, it could capture meaningful share of the $50B+ AI compute market before Google Cloud's playbook repeats, and even 8-12% margins on $20B+ revenue would justify the capex.
"Meta's entry into cloud will dilute its 41% operating margins without offsetting revenue growth that justifies the $145B capex trajectory."
Meta's cloud push, framed as monetizing excess AI capacity, directly addresses capex skepticism after raising 2026 guidance to $145B. Yet cloud's typical 18% operating margins, as seen at Google, sit far below Meta's 41% and 82% gross margins from ads. Building enterprise sales teams and competing with CoreWeave-style neoclouds adds costs Meta has never carried. The 9% rally on the news overlooks how long it took peers to reach profitability and risks normalizing lower returns on infrastructure built primarily for internal use.
Meta's existing data-center scale and Nvidia relationships could let it undercut rivals on price while still clearing 25-30% margins, turning capex into a genuine high-ROIC business faster than the article assumes.
"Meta's cloud push risks near-term margin compression and capital intensity, with meaningful upside conditional on achieving rapid scale and strong cross-sell dynamics that may not materialize quickly."
Meta’s cloud pivot is framed as a diversification into a high-margin, scalable business, but the near-term math looks painful. Cloud margins at peers sit well below Meta’s 41% operating margin, and Meta would face front-loaded capex, enterprise sales costs, energy bills, and potential time-to-revenue friction. The piece glosses over Google’s cloud margin (~18%) and underplays how long it takes to monetize external compute at scale. If external demand stalls, Meta could incur heavy depreciation and weak ROIC while the ad business bears the burden of funding the capital cycle.
But Meta’s scale and potential cross-sell to advertisers could improve overall margins faster than expected if external compute demand accelerates and data-center efficiency compounds.
"Meta's cloud pivot creates a massive antitrust liability regarding self-preferencing that the market is currently ignoring."
Claude, you’re missing the regulatory angle. By positioning as a cloud provider, Meta invites antitrust scrutiny into its internal compute allocation. If Meta uses its 'spare' capacity to train its own Llama models while charging competitors premium rates, they trigger a 'self-preferencing' investigation similar to the DMA challenges Google faces in Europe. This isn't just a margin compression risk; it’s a strategic trap that could force Meta to divest infrastructure or face debilitating oversight, far outweighing the short-term revenue gains.
"Regulatory risk is real but structurally containable; the real lever is whether cloud revenue exceeds the opportunity cost of idle capacity depreciation."
Gemini's regulatory trap is real, but underweights Meta's escape hatch: they can ring-fence external cloud ops under a separate legal entity with transparent pricing audits, mirroring how Google structures YouTube vs. Search. The DMA risk exists, but it's manageable through governance, not disqualifying. What nobody's flagged: Meta's $145B capex is already sunk. The question isn't whether cloud is profitable in isolation—it's whether 12-15% margins on $15-20B revenue beats the depreciation hit of idle GPUs. That math still favors monetization.
"Shared infrastructure makes ring-fencing ineffective against self-preferencing claims."
Claude's ring-fence via separate entity ignores that Llama training runs on the identical GPU fabric Meta would rent out. Any audit would expose priority queuing favoring internal models, recreating the exact DMA self-preferencing exposure Gemini flagged. This structural overlap caps credible external revenue far below the $15-20B needed to justify idle-capacity monetization against ongoing $145B capex.
"External compute revenue won't reach 25-30% margins because DMA risk and cross-tenant constraints cap pricing, forcing Meta to rely on capex-heavy scale instead of a high-ROIC outsourcing play."
Grok, I’d push back on the ‘25-30% external margin’ thesis. Even with isolated multi-tenant controls, DMA-style scrutiny will constrain both pricing and prioritization. Internal workloads will always have implicit subsidies, and external clients face performance risks that deter large, durable contracts. Net-net: external compute revenue likely stays far below internal margin potential, forcing Meta to rely on capex-heavy scale-up and cross-sell, not a clean, high-ROIC outsourcing play.
The panelists generally agree that Meta's entry into the cloud infrastructure space is risky, with potential margin compression and regulatory challenges outweighing short-term revenue gains. They express concern about Meta's ability to differentiate, monetize excess capacity, and avoid antitrust scrutiny.
Monetizing excess AI capacity to justify ongoing capex, although panelists express skepticism about the feasibility and profitability of this strategy.
Regulatory scrutiny and potential 'self-preferencing' investigations due to Meta's internal compute allocation and prioritization of its own models.