Is Meta Entering AI Cloud Services? Watch Compute Sales, Broadcom Chips, and Margins Before Earnings

Meta entering AI cloud services sounds like a new growth story on the surface, but it is really a return-on-investment test under AI capital spending pressure. Meta has already built massive compute infrastructure for ad recommendations, generative AI, Meta AI, Llama, smart glasses, and future application interfaces. If part of that compute can be rented externally, or if model access can be packaged into developer services, the market will naturally reassess Meta's AI investment payback period.

But this should not be simplified into "Meta is copying AWS, Azure, or Google Cloud." Meta's advantage is not a traditional enterprise cloud sales system. It lies in its enormous internal AI workloads, advertising and content recommendation scenarios, Llama ecosystem, and MTIA custom AI chips being developed with Broadcom. The real pre-earnings question is whether selling AI compute would improve asset utilization and margins, or push Meta into even heavier AI infrastructure expansion.

Key Takeaways

• Meta will report Q2 earnings after the U.S. market close on July 29, 2026, with the conference call held the same day.
• Bloomberg has reported that Meta is studying the sale of AI compute and model access, but this should still be treated as a potential business direction.
• Meta and Broadcom have announced an expanded MTIA custom AI chip partnership. Broadcom mentioned a 2nm AI accelerator, multi-gigawatt deployments, and a roadmap extending to 2029.
• Reuters reported that Meta plans to begin producing a new AI chip in September and raise total compute capacity to 14GW.
• Meta's 2026 CapEx guidance has already risen to $125 billion to $145 billion, so whether compute monetization can improve investment returns will be a key earnings-call issue.
• After earnings, Meta stock needs to be evaluated through advertising revenue, AI spending, Broadcom chip progress, potential AI cloud service commentary, and operating margins together.

Is Meta Really Entering AI Cloud Services: A Confirmed Business or a New Pre-Earnings Narrative?

Meta AI cloud services are better understood for now as a new variable that needs earnings validation, not as a mature new business line that has already been fully disclosed. Bloomberg reported in early July that Meta was developing plans for a cloud infrastructure business that could sell AI computing power and model access to external customers, turning excess compute into revenue. Bloomberg later reported that Meta had held early talks with Anthropic about renting data center compute, with a potential arrangement that could reach the $10 billion range. Because this information mainly comes from media reporting and people familiar with the matter, management confirmation on the earnings call will be crucial.

More specifically, what Meta may be exploring is probably not a full general-purpose cloud business. AWS, Azure, and Google Cloud cover compute, storage, databases, security, enterprise applications, networking, developer tools, and industry solutions. Meta's potential entry point is more likely to be AI compute, model hosting, API access, and Llama or Muse Spark model services. In other words, if Meta enters AI cloud, it would likely start with "compute and model services," not instantly become the fourth comprehensive cloud platform.

Zuckerberg's comments in a Bloomberg interview also fit this framework. He said market pricing for compute is high enough that, in some cases, renting capacity or making similar arrangements could make sense. That is not formal revenue guidance, but it shows Meta's management is comparing the opportunity cost of using compute internally versus renting it externally. For investors, this is more important than the broad question of whether Meta wants to do cloud, because it connects directly to AI capital spending returns.

The earnings call should be listened to for three things: whether Meta acknowledges external compute monetization, whether the revenue model would be compute rental, model API, or strategic partnership, and whether such revenue would be disclosed separately in the future. If management only responds with broad language such as "we evaluate opportunities," the market may still treat AI cloud as an option rather than a defined line item in valuation models.

Summary: Meta AI cloud services currently look more like a pre-earnings narrative that needs validation. Investors should distinguish between "renting some excess compute" and "building a full cloud platform." The former is an asset-utilization issue; the latter is a cloud-business strategy.

Why Would Meta Consider Selling Compute: Is AI CapEx Too High, or Has Capacity Planning Created Redundancy?

The backdrop for Meta potentially selling compute is that capital spending has become large enough to require clearer explanation. Meta's Q1 2026 earnings report showed Q1 capital expenditures, including principal payments on finance leases, of $19.84 billion. The company also raised 2026 CapEx guidance from $115 billion to $135 billion to $125 billion to $145 billion, citing higher component prices and additional data center costs to support capacity in future years.

This investment intensity is subtly changing Meta's business model. In the past, the market viewed Meta as an asset-light advertising platform, with user growth, ad pricing, content recommendations, and regulatory risk as the core issues. Now Meta is being reclassified into the comparison set of AI infrastructure giants. Microsoft, Amazon, and Alphabet can absorb external AI demand through Azure, AWS, and Google Cloud. Meta has historically relied mainly on advertising cash flow to digest infrastructure investment. If Meta can sell part of its compute to external customers, it would add a second revenue outlet for AI CapEx.

However, "selling compute" can mean two very different things. The first is short-term capacity mismatch: Meta builds quickly, and certain regions, time windows, or hardware types have rentable capacity; external sales improve asset utilization. The second is strategic expansion: Meta deliberately plans larger capacity for external customers and treats AI compute as a long-term business. The former is more like cash-flow optimization, while the latter is a new business investment with greater risk and valuation implications.

That is why Meta's CapEx commentary in earnings will be so important. If the company maintains the $125 billion to $145 billion range and explains compute sales as a way to improve existing capacity utilization, the market may see it as helpful in reducing AI investment anxiety. If Meta raises CapEx again and cites external compute demand as a reason, investors may instead worry that the company is entering a heavier, more cyclical, and less visible infrastructure competition.

Compared with cloud vendors, Meta has clear strengths and weaknesses. Its strength is that internal AI workloads are large enough to first optimize infrastructure for its own ads, recommendations, and AI products before opening available capacity externally. Its weakness is that enterprise cloud sales, customer support, service-level agreements, data isolation, and compliance are not historically core parts of Meta's business model. Therefore, for compute sales to become high-quality revenue, demand scarcity alone is not enough. Meta also needs long-term productization ability.

Summary: Meta's compute-sales logic is not simply a transition into being a cloud vendor. It is about giving massive AI infrastructure a second revenue outlet. Whether that improves valuation depends on whether Meta is selling existing idle capacity or building a new infrastructure business that requires even more CapEx.

Why Broadcom Chips Matter: Can MTIA Lower Meta's AI Inference Costs?

Broadcom and MTIA are central to understanding the margin potential of Meta AI cloud services. In its MTIA roadmap update, Meta said it will develop and deploy four new generations of MTIA chips over two years for ranking, recommendations, and generative AI workloads. Meta also emphasized an inference-first strategy, aiming for better cost efficiency than general-purpose chips in workloads that fit its own needs.

Meta then announced an expanded partnership with Broadcom. Meta's official partnership announcement said the two companies will co-develop multiple generations of next-generation MTIA chips. Broadcom's press release added that the partnership will support Meta's multi-gigawatt custom silicon deployment, with an initial commitment of more than 1GW, an industry-first 2nm AI compute accelerator, and a plan extending to 2029. This indicates that MTIA is not just an experiment, but a core part of Meta's AI infrastructure.

Reuters, in a report republished by Investing.com, said Meta plans to start producing an AI chip code-named Iris in September and raise total compute capacity to 14GW next year. The report also said Iris is part of the four-generation MTIA project and is intended to strengthen AI capabilities behind platforms such as Facebook and Instagram. If management reinforces these developments on the earnings call, the market will be more willing to believe Meta can control AI inference costs.

The most important role of custom silicon is not that "Meta can also make chips," but that it can change unit economics. AI cloud gross margin depends on chip cost, power cost, network efficiency, utilization, depreciation life, and pricing power. If Meta mainly relies on external GPUs, it can rent compute at high prices when capacity is scarce, but prices may fall when supply improves. If MTIA can consistently reduce costs in recommendations, ads, inference, and some generative AI workloads, Meta could achieve more stable margins across internal applications and external services.

Of course, MTIA is not a cure-all. Custom chips require design, validation, manufacturing, packaging, networking, software stack support, and operations to work together. Delays in any part can affect deployment timing. More importantly, whether external customers are willing to use Meta's custom compute depends on software compatibility, model ecosystem, API usability, performance stability, and service commitments. If external customers prefer the NVIDIA ecosystem or mature cloud platforms, Meta will need time to turn low-cost chips into external revenue.

Summary: The importance of Broadcom and MTIA is not the chip narrative itself, but whether Meta can use custom hardware to reduce inference costs. Only if the cost curve improves can compute sales, model APIs, and AI application expansion support margins.

Will AI Cloud Services Raise or Lower Margins: How Should Investors Measure Revenue, Depreciation, and Utilization?

AI cloud services do not automatically improve Meta's margins. Compute businesses may add new revenue, but they are also asset-heavy, depreciation-heavy, power-intensive, network-intensive, and highly cyclical in supply and demand. For a platform company whose core profit source has historically been advertising, external compute revenue with insufficient gross margin could dilute group margins or make investors more worried about free cash flow.

The most important variable is utilization. Suppose Meta has already built data centers and chip clusters for internal AI workloads, but part of its capacity is temporarily underused. In that case, external rental can turn sunk costs into incremental revenue, and marginal profit may be attractive. Conversely, if Meta builds extra capacity specifically for external customers, revenue increases, but CapEx, depreciation, power purchases, operations, and finance leases also increase. Margin improvement then becomes much less certain.

The second variable is pricing power. AI compute is currently scarce, and external customers are willing to pay high prices for GPU clusters, inference capacity, and model hosting. But compute pricing does not only move upward. As NVIDIA, AMD, custom ASICs, cloud vendors' in-house chips, and more data center supply come online, compute rental prices may enter a more competitive phase. If Meta lacks software ecosystem, model APIs, developer tools, and enterprise service capabilities, hardware capacity alone may not sustain high gross margin.

Investors also need to distinguish between advertising margins and AI cloud margins. Meta's Family of Apps business has strong operating leverage, with Q1 Family of Apps income from operations reaching $26.9 billion. Reality Labs, by contrast, lost $4.028 billion. If AI cloud revenue is added in the future, investors need to know which segment it belongs to, what its gross margin is, whether it uses the same AI assets, and whether external allocation creates an opportunity cost for ads and Meta AI.

If management is willing to disclose capacity utilization, cost per inference, external compute revenue, model API usage, or AI infrastructure payback, the market will have an easier time incorporating AI cloud into valuation. If disclosure remains at the level of "demand is strong and the opportunity is large," short-term trading is more likely to revolve around CapEx increases and margin compression.

Summary: AI cloud services can improve Meta's margins only if they raise existing compute utilization, lower unit inference cost, and create sustainable external demand. If they require additional data center expansion, they may increase free-cash-flow pressure in the short term.

Can Advertising Still Cover AI Spending: What to Watch in Q2 Revenue, Ad Pricing, and Operating Margins?

Meta's AI cloud narrative ultimately returns to advertising cash flow. In Q1 2026, revenue was $56.311 billion, up 33% year over year; ad revenue was $55.024 billion, up 33%; ad impressions rose 19%, and average price per ad rose 12%. These numbers show that Meta's core advertising business is still supported by both volume and pricing. Without that advertising cash-flow layer, AI CapEx, MTIA chips, and potential compute sales would be harder for the market to tolerate.

The Q2 revenue threshold is already clear. Meta's official Q2 revenue guidance is $58 billion to $61 billion. Investopedia reported that analysts expect Q2 revenue of about $60.23 billion and EPS of about $7.19. Benzinga shows expectations of roughly $60.22 billion in revenue and EPS of $7.18. If Q2 revenue merely meets expectations, AI cloud rumors alone may not lift valuation. If revenue approaches the upper end of the range and ad pricing continues to rise, the market will be more willing to accept Meta's infrastructure spending.

Ad pricing is especially important. Impression growth can come from more content distribution and more advertising inventory, but rising ad prices usually mean advertiser demand is strong and campaign ROI is still improving. If Q2 average price per ad keeps rising, Meta's recommendation systems, creative tools, ad automation, and conversion efficiency are still working. In that case, AI spending is not only being used for future vision; it is also producing returns in the current advertising system.

If advertising is strong, Meta can frame AI cloud and Broadcom chips as "a second curve built on strong cash flow." If advertising slows, compute sales may be interpreted by the market as "a way to find an outlet for excessive CapEx." This is why the same news can trigger different stock reactions under different fundamental conditions. What the market really wants to see is a closed loop among ad growth, AI spending, chip-driven cost reduction, and external compute revenue.

Reality Labs also cannot be ignored. Q1 Reality Labs revenue was only $402 million, while its operating loss reached $4.028 billion. If Q2 losses widen and AI cloud services have no clear revenue contribution, group operating margins will face more questions. If management can connect Reality Labs, AI glasses, Meta AI, and compute infrastructure while providing clearer commercialization metrics, long-term narrative value may partially offset the loss pressure.

Summary: Meta's AI cloud narrative is ultimately supported by advertising cash flow. If Q2 ad revenue, ad pricing, and Family of Apps margins remain strong, the market will give Meta more time to prove AI infrastructure returns. If advertising weakens, compute sales may be viewed as a defensive fix.

What to Watch When Trading Meta Stock After Earnings: Biya Checklist and Risk Control

Trading Meta stock after earnings should not be based only on the "AI cloud services" concept. A more complete framework should include four variables: whether advertising revenue beats expectations, whether AI CapEx is raised again, whether Broadcom/MTIA provides cost-reduction clues, and whether management confirms external compute monetization. If all four variables lean positive, the stock is more likely to find valuation support. If advertising only meets expectations while CapEx continues to rise, AI cloud rumors could amplify volatility instead.

Short-term reactions also need to be read against after-hours liquidity. Investopedia reported that options-market pricing before Meta earnings implied a roughly 7% two-way move in the stock. After-hours quotes can move quickly because of thin liquidity, wider bid-ask spreads, and algorithmic trading. Ordinary investors are better off reviewing the full report and earnings-call takeaways before deciding whether to trade, rather than reacting only to the first after-hours candle.

For users who need to watch or trade U.S. stocks, BiyaPay can be used to follow Meta, Broadcom, Microsoft, Alphabet, Amazon, and other U.S. stock names, while combining earnings schedules, order types, and personal risk tolerance into a plan. BiyaPay supports multi-asset trading across U.S. stocks, Hong Kong stocks, and cryptocurrencies. Users can also view tradable names through the U.S. stock list, or use mobile services through the app download page.

On fees, BiyaPay's U.S. stock commission is $0. Platform fees, external institution fees, FX costs, order prices, and other applicable fees should be based on the BiyaPay pricing page, order-page display, and platform rules. Post-earnings trading also requires attention to the difference between limit orders and market orders, after-hours market depth, bid-ask spreads, applicable rules across different accounts and regions, and the maximum drawdown an individual can tolerate.

A more disciplined trading sequence is to first see whether headline results beat consensus, then examine ad volume-price structure and operating margins, then listen to management's explanation of AI cloud, Broadcom, MTIA, and CapEx, and finally choose order types based on after-hours and next-day opening liquidity. The biggest risk on earnings day is not misreading one number, but treating an unconfirmed commercialization narrative as already-realized profit.

Summary: The focus after Meta earnings is not chasing the "AI cloud" concept, but judging whether the market believes Meta can connect advertising cash flow, Broadcom custom chips, and external compute revenue into a verifiable margin path. The more confirmation there is, the stronger valuation support becomes. The more vague the disclosure is, the more easily volatility can expand.

FAQ

Has Meta Officially Launched AI Cloud Services?

For now, it is better understood as a potential business direction discussed in media reports and management commentary. Public information indicates that Meta is studying the sale of AI compute and model access, but whether it becomes a formal business line, how revenue would be disclosed, and which customers it would serve still require company confirmation or earnings-call commentary.

How Is Meta Selling AI Compute Different From AWS or Azure?

AWS, Azure, and Google Cloud are full cloud platforms covering compute, storage, databases, security, enterprise services, and developer ecosystems. Meta's potential entry point is more likely to be AI compute, model hosting, model APIs, and developer access. The scope is narrower and more dependent on Llama, Muse Spark, and Meta's internal AI infrastructure.

What Role Does Broadcom Play in Meta's AI Chips?

Broadcom is working with Meta on multiple generations of MTIA custom chips and supporting Meta's multi-gigawatt custom silicon deployment. Its role includes chip design, packaging, networking, and XPU platform capabilities, helping Meta optimize the performance and cost of AI inference, recommendation, and generative AI workloads.

Will AI Cloud Services Improve Meta's Margins?

Not necessarily. If Meta only rents existing idle capacity, it may improve asset utilization and cash recovery. If it keeps building data centers for external customers, it could continue to push up CapEx, depreciation, and free-cash-flow pressure in the short term. The key variables are utilization, pricing power, MTIA cost, and customer-demand stability.

What Are the Most Important Metrics in Meta's Q2 Earnings?

The most important metrics are whether revenue approaches the $61 billion upper end of guidance, whether ad pricing keeps rising, whether CapEx stays within the $125 billion to $145 billion range, whether Family of Apps margins remain stable, and whether management confirms a compute monetization plan.

What Fees Should Investors Watch When Trading Meta Stock After Earnings?

Investors should watch commissions, platform fees, external institution fees, FX costs, bid-ask spreads, after-hours liquidity, and order types. BiyaPay's U.S. stock commission is $0, but other fees and applicable rules should be based on the pricing page, order display, and platform rules.