Wire flash
Goldman Sachs: US Top 5 Cloud Giants' AI Infrastructure Spending to Surge Over 50% Next Year to $1.2 Trillion
Editorial responsibility
- No named human review is recorded for this page.
- Source reporting is collected, normalized, translated or condensed automatically when needed.
- Automatically published source-backed update
Goldman Sachs strategists forecast that the five largest US hyperscale cloud providers will increase their AI infrastructure spending by over 50% next year, reaching $1.2 trillion. The analysis shifts the debate from the scale of spending to the potential returns on this massive investment. Goldman outlines a three-phase AI buildout: Phase 1 (2023-2025) at $633 billion, Phase 2 (2026-2027) at $1.73 trillion, and Phase 3 (2028-2030) at $4.14 trillion. To achieve a 15% return on invested capital (ROIC) on the Phase 2 spending, the six major hyperscalers (Alphabet, Amazon, Microsoft, Meta, Oracle, SpaceX) would need to generate approximately $1.42 trillion in cumulative revenue from 2028 to 2030. The report highlights a key risk from depreciation, which could be significantly higher if GPU lifespans are shorter than the assumed 5 years. Cloud backlog orders of $1.69 trillion provide some support for the revenue targets, but the analysis also notes risks from off-balance-sheet commitments and the massive financing gap for the overall spending, estimated by Morgan Stanley at $1.5 trillion. Goldman maintains a 'buy' rating on the six firms, viewing near-term return compression as a natural result of a large upfront investment cycle.
Source report
Goldman Sachs strategists project that capital expenditure by the five major U.S. hyperscale cloud providers on AI infrastructure will rise by more than 50% next year, reaching $1.2 trillion.
The AI capital expenditure boom among hyperscalers is shifting from a debate over "how much to spend" to the core question of "how much can be earned back."
According to the latest Goldman Sachs research, for the six major hyperscale cloud companies to achieve a 15% return on invested capital (ROIC) on approximately $1.73 trillion in AI computing capital expenditure between 2026 and 2027, they would need to generate cumulative revenue of roughly $1.42 trillion between 2028 and 2030.
Eric Sheridan, head of Goldman Sachs' Internet and Technology team, noted in the report that the market debate has shifted from the scale of near-term spending to whether these massive investments can generate meaningful returns over the next three years. Meanwhile, cloud backlog data provides some support: as of Q2 2026, the combined backlog for AWS, Azure, and Google Cloud stood at approximately $1.69 trillion, up 152% year-over-year, offering preliminary validation of revenue target feasibility.
However, variables outside Goldman's framework cannot be ignored. Approximately $3 trillion in off-balance-sheet commitments are not included in the ROIC denominator. Downward pressure on GPU and computing pricing has already emerged, and the return question for the even larger "Phase 3" capital expenditure (approximately $4.14 trillion from 2028 to 2030) is explicitly set aside for separate discussion.
Capital Expenditure in Three Phases: Escalating Scale
Goldman Sachs divides the current AI computing buildout cycle into three phases:
- Phase 1 (2023–2025): Total capital expenditure of approximately $633 billion, averaging ~$211 billion annually, primarily for training infrastructure for a small number of foundation model companies.
- Phase 2 (2026–2027): Total capital expenditure of approximately $1.73 trillion, averaging ~$863 billion annually, as demand extends to inference and computing supply remains tight.
- Phase 3 (2028–2030): Total capital expenditure of approximately $4.14 trillion, averaging ~$1.38 trillion annually, with capital intensity expected to moderate marginally as hyperscalers potentially enter "harvest mode."
Notably, since the beginning of this year, consensus estimates for 2026–2027 capital expenditure among the five publicly listed hyperscalers (Alphabet, Microsoft, Amazon, Meta, and Oracle) have been revised upward by approximately 66%, a total increase of roughly $754 billion, from $1.14 trillion to $1.89 trillion. Goldman Sachs states that its own 2027 capital expenditure forecast remains above market consensus and is "more aligned with the actual expectations of buy-side investors."
From Spending to Revenue: Goldman's Calculation Framework
Goldman's calculation logic is clear: of the $1.73 trillion in Phase 2 capital expenditure, approximately 70% ($1.21 trillion) is allocated to computing (servers, chips, networking), and approximately 30% ($518 billion) to infrastructure "shell" (land and buildings). Among the six companies, Alphabet has the highest capital expenditure ($470 billion), followed by Amazon ($424 billion), Microsoft ($327 billion), Meta ($309 billion), Oracle ($165 billion), and SpaceX ($31 billion).
Key assumptions include a cost of approximately $42 billion per gigawatt of computing capacity, a 5-year depreciation period for computing assets, 15 years for infrastructure shell, annual operating costs of approximately $836 million per gigawatt, and utilization rates of 60%, 80%, and 85% for 2028, 2029, and 2030, respectively.
Under the 15% ROIC target, the six companies would need to generate approximately $129.5 billion in annual net operating profit after tax (NOPAT), corresponding to roughly $164 billion in annual earnings before interest and taxes (EBIT). Adding approximately $276 billion in annual depreciation and amortization for Phase 2 assets yields required annual revenue of approximately $465 billion to $476 billion, totaling roughly $1.42 trillion over three years, implying an EBIT margin of approximately 35%.
Depreciation Pressure: The Underestimated Core Risk
Perhaps the most striking figure in this calculation is not the revenue target itself, but the scale of depreciation—annual depreciation and amortization for Phase 2 assets alone reaches $276 billion, approximately 1.7 times the EBIT these assets need to generate.
This means the economics of the current AI investment cycle depend heavily on the actual useful life of GPUs. Goldman assumes a 5-year depreciation period. If the actual economic life is only 3 years (as some skeptical analysts and Nvidia's own annual product iteration cadence suggest), hyperscalers would be forced to purchase next-generation chips at a higher frequency, significantly raising the revenue threshold.
Goldman's sensitivity analysis shows that with ROIC targets ranging from 0% to 30% and capital expenditure per gigawatt ranging from approximately $34 billion to $51 billion, the cumulative revenue required for the six companies from 2028 to 2030 falls between approximately $900 billion and $1.9 trillion, translating to annual revenue per gigawatt of approximately $6.2 billion to $18.6 billion. Even in a zero-return scenario, covering only depreciation and operating costs would require $920 billion in revenue.
Cloud Backlog: The Most Important "Safety Net"
Goldman's primary evidence supporting the feasibility of the revenue target is cloud backlog data. As of Q2 2026, the combined backlog for AWS, Azure, and Google Cloud stood at approximately $1.69 trillion, roughly 1.5 times higher than at the start of the year, with public cloud revenue growth accelerating to over 45% year-over-year.
For these three companies alone, their approximately $1.22 trillion in Phase 2 capital expenditure corresponds to roughly $1 trillion in revenue requirements for 2028–2030, equivalent to approximately 59% of the current backlog—and this calculation assumes zero backlog growth thereafter. By company, the required revenue represents approximately 40% of Microsoft's backlog, 68% of Amazon's, and 78% of Alphabet's.
Goldman characterizes this assumption as "potentially conservative." However, the report also notes that the backlog data includes non-cloud business commitments and is not strictly composed of purchase orders. Additionally, a significant portion of orders comes from a small number of AI labs whose payment capacity depends heavily on continued access to capital markets, posing customer concentration risk.
Morgan Stanley's Perspective: The Financing Gap Is Also Vast
Comparing Goldman's revenue threshold with Morgan Stanley's research provides a more complete picture. Morgan Stanley estimates that GenAI investments can achieve incremental ROIC of 25% to 50%—approximately 31% for GPU leasing, 46% for model APIs based on owned infrastructure, and 25% for third-party infrastructure. Morgan Stanley assumes annual GPU leasing revenue of approximately $22.9 billion per gigawatt and model API revenue of approximately $30.4 billion per gigawatt, both above the $11.6 billion per gigawatt implied by Goldman's 15% ROIC target. The two firms broadly agree: if GPU leasing prices and computing pricing remain near current levels, the revenue targets are mathematically achievable.
However, Morgan Stanley's earlier research highlights financing pressure: it estimates global data center capital expenditure of approximately $2.9 trillion from 2025 to 2028, with hyperscalers' own cash flow covering only about $1.4 trillion. The remaining financing gap of approximately $1.5 trillion would need to be filled by roughly $800 billion in private credit, $200 billion in corporate bonds, $150 billion in ABS/CMBS, and $350 billion from other sources. Goldman's latest forecast, with combined Phase 2 and Phase 3 capital expenditure of approximately $5.9 trillion, far exceeds these estimates.
Blind Spots in Goldman's Framework: Off-Balance-Sheet Commitments and Phase 3
Goldman acknowledges the limitations of its own framework. Its ROIC calculation explicitly excludes off-balance-sheet lease obligations, finance leases, special purpose vehicles (SPVs), and third-party computing licensing arrangements.
This is not a minor footnote. According to relevant data, off-balance-sheet commitments in the AI sector have reached approximately $3 trillion and are growing rapidly. These commitments also need to generate returns or at least be serviced, and credit markets have already begun to react.
Furthermore, Goldman's framework covers only Phase 2. Applying the same revenue-to-capital-expenditure ratio (approximately 0.82x) roughly to the approximately $4.14 trillion Phase 3 forecast would require hyperscalers to generate an additional $3.4 trillion in revenue on top of the $1.42 trillion Phase 2 target. Peter Berezin of BCA Research goes further, suggesting that if data center spending continues at current levels, it may require up to $10 trillion in annual AI revenue to monetize all capital expenditure.
Goldman maintains a "Buy" rating on all six companies and characterizes near-term return compression as "a natural result of large-scale upfront investment cycles, not evidence of structurally poor AI economics." However, this statement holds true within both bullish and bearish narrative frameworks.
Risk Disclaimer: Market conditions involve risks, and investment decisions should be made with caution. This article does not constitute personal investment advice and does not account for individual users' specific investment objectives, financial situations, or needs. Users should consider whether any opinions, views, or conclusions herein are suitable for their particular circumstances. Investment based on this content is at the reader's own risk.
Source
华尔街见闻Neutral / independent
Part of this Story
Goldman Sachs: US hyperscalers need $1.42 trillion revenue by 2030 for AI ROI