Meta's potential AI cloud business provides a “silver lining” to capital expenditure concerns

Overview

This report centers on Bloomberg's statement that Meta is evaluating a cloud infrastructure business, with the core idea being to sell AI compute capacity and hosted model access to external customers. Deutsche Bank believes this direction could establish a more direct revenue bridge for Meta's currently large-scale AI capital expenditures and improve the narrative around returns on AI infrastructure investment.

Core views

The report's core views include: first, potential third-party compute sales do not mean Meta is exiting frontier AI, and Meta still regards superintelligence and large-scale training clusters as strategic priorities; second, capital expenditures may remain elevated in the short term, but if there is a direct third-party revenue stream, market concerns about idle capacity and capital intensity would decline; third, if 75% of the 1.2 to 2.65GW of salable compute is leased or sold in FY27, it could generate meaningful revenue and EPS upside; fourth, model API or Bedrock-like business revenue is of higher quality than raw compute leasing, but commercialization requires higher execution capability.

Analysis framework

The report uses a combination of scenario analysis and peer benchmarking: it first estimates Meta's total AI-related capacity by the end of FY27, then sets assumptions for the salable proportion, sell-through rate, and annualized revenue per GW for new capacity added in different years, and subsequently derives the impacts on revenue, operating margin, after-tax net income, and EPS, using AWS, Google Cloud, CoreWeave, and compute companies such as Nebius and CoreWeave, along with relevant contracts, as references for pricing and margins.

Methodology notes

• Scenario analysisFY27E compute sale scenarios

  Estimate Meta's salable AI compute, sell-through rate, and unit revenue under scenarios A, B, and C.

  The report assumes Meta's total AI-related capacity at the end of FY27 will be 8.0 to 11.5GW, of which 1.2 to 2.65GW could become salable capacity, and assumes that 75% of the salable capacity is actually sold.

• Incremental profit analysisContribution margin framework

  For underutilized capacity within already-committed infrastructure, the focus is on incremental contribution margin rather than the mature-stage margin of a full cloud business.

  The report assumes an operating margin of 50% to 75% on incremental revenue because depreciation, data center, and part of the operating costs have already been absorbed by market expectations.

• Peer benchmarkingCloud and AI compute service benchmarking

  Use AWS, Google Cloud, CoreWeave, and compute contracts as references for margin and pricing assumptions.

  AWS's 2025 operating margin is about 35%, Google Cloud's current margin is about 24%, and CoreWeave's 2025 adjusted EBITDA margin is about 60%; if Meta resells already-committed capacity in the short term, the contribution margin could be relatively high.

Asset mapping & comparison

Structured mapping from thesis to named assets (strengths, weaknesses, peers, risks).

• META.US / META PLATFORMS INC

  Core coverage target, with potential to benefit from the commercialization of AI compute and model APIs.

  Strengths

  It already has large-scale AI infrastructure investment; if it sells non-core or older capacity, revenue flow-through could be high and idle-capacity risk could be mitigated.

  Weaknesses

  Capital expenditures remain high, and its enterprise cloud distribution capability, customer support, and compliance system are weaker than those of AWS, Azure, and GCP.

  Comparison

  Compared with mature cloud providers, Meta lacks enterprise cloud channels; compared with AI compute companies such as CoreWeave, Meta's advantage lies in its existing large-scale internal infrastructure and model ecosystem.

  Risks

  Salable capacity, pricing, customer demand, execution capability, and internal training priorities could all come in below the scenario assumptions.

• CRWV.US / COREWEAVE INC

  AI compute service benchmark company, used as a reference for revenue per unit of capacity and margin.

  Strengths

  The market already views it as a major beneficiary in an environment of tight AI compute supply.

  Weaknesses

  GAAP profitability remains weighed down by depreciation, interest, and growth investments.

  Comparison

  The report uses CoreWeave's revenue guidance, capacity targets, and approximately 60% adjusted EBITDA margin as one of the references for Meta's potential compute sales.

  Risks

  If AI compute supply constraints ease, industry pricing and margins could both decline.

• NBIS.US / NEBIUS GROUP NV

  Reference target for AI infrastructure pricing and capacity assumptions.

  Strengths

  Benefits from the growth in external demand for AI compute.

  Weaknesses

  Its business scale and customer mix differ from Meta's, limiting direct comparability.

  Comparison

  The report uses Nebius's and CoreWeave's 2026 revenue guidance and capacity targets for low-end pricing assumptions.

  Risks

  Its valuation and pricing reference may be affected by cyclicality in AI infrastructure.

Key data

• AI-related capacity by the end of FY27Approximately 8.0 to 11.5GWBased on assumptions of about 2GW capacity by the end of FY25, about 2.0 to 3.5GW added in FY26, and about 4.0 to 6.0GW added in FY27.
• Potential salable computeApproximately 1.2 to 2.65GWDerived from older, non-core, or temporarily underutilized compute capacity, while the latest capacity is still expected to be prioritized for internal frontier model training.
• FY27 incremental revenue opportunityApproximately USD 9 billion to USD 29.8 billionEquivalent to 3.0% to 9.9% above the Street FY27 revenue consensus of about USD 301.8 billion, with the base-case scenario at about USD 17.5 billion.
• FY27 operating margin improvementApproximately 44 to 361 basis pointsUnder the assumption of 50% to 75% incremental profit margins, potential operating margin rises from 34.8% to 35.2% to 38.4%.
• Potential FY27 GAAP EPS upsideApproximately 4.0% to 21.3%Potential EPS of about USD 36.51 to USD 42.59, above the Street consensus of USD 35.12; base-case upside is about 10.2%.
• Recent priceUSD 563.29The disclosure list shows the price date as 2026-06-30.

Impact & implications

If Meta successfully establishes a third-party AI compute or model API revenue stream, investors may no longer view AI capital expenditures solely as indirect investment in advertising, recommendations, and future products, but instead as infrastructure assets capable of generating direct external revenue. This could improve the market's view of Meta's earnings quality, capital expenditure risk, and valuation multiple, though it does not imply lower capital expenditures in the short term.

Risks

• Meta may prioritize its latest chips and highest-quality capacity for internal superintelligence training, resulting in lower salable compute than assumed.
• If AI compute supply-demand tightness eases, annualized revenue per GW and sell-through rates may come in below the report's scenarios.
• A third-party cloud business requires capabilities in enterprise sales, SLA, billing, procurement, compliance, security, developer tools, model governance, and customer support; Meta currently does not have a distribution foundation comparable to AWS, Azure, or GCP.
• Capital expenditures may still remain elevated, and a potential revenue stream does not mean a short-term reduction in total capital expenditures.
• If selling capacity affects internal model progress or advertising product improvements, it could weaken long-term strategic value.
• The report discloses that Deutsche Bank and its affiliates had non-investment-banking service relationships and related compensation arrangements with Meta over the past year, and investors should pay attention to the disclosure of potential conflicts of interest.

What to watch

• Whether Meta formally announces a cloud infrastructure or model API business, and whether the product takes the form of raw compute leasing or a Bedrock-like API.
• FY26 and FY27 AI capital expenditure guidance, progress in GW-scale capacity buildout, and the rollout pace of clusters such as Prometheus and Hyperion.
• The chip generation, use cases, and whether salable capacity conflicts with internal frontier model training.
• The duration, pricing, sell-through rate, and revenue recognition method of third-party customer contracts.
• Progress in building the SLA, billing, compliance, security, model governance, and customer support capabilities required for the cloud business.
• Whether Street expectations for FY27 revenue, operating profit, and EPS are revised upward, and whether the market assigns a higher valuation multiple.