The AI capital expenditure frenzy of hyperscale cloud providers 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, the six major hyperscale cloud providers would need to generate approximately $1.42 trillion in cumulative revenue during 2028 to 2030 in order to achieve a 15% return on invested capital (ROIC) on roughly $1.73 trillion in AI computing capital expenditure during 2026 to 2027.
Eric Sheridan, head of Goldman Sachs' internet and technology team, noted in the report that the current market debate has shifted from the scale of near-term spending itself to whether this massive investment round can generate meaningful returns within the next three years. Meanwhile, cloud backlog data provides some support 鈥?as of the second quarter of 2026, AWS, Azure, and Google Cloud together had a backlog of approximately $1.69 trillion, up 152% year over year, offering preliminary validation for the feasibility of the revenue targets.
However, variables outside Goldman Sachs' framework also cannot be ignored. Roughly $3 trillion in off-balance-sheet commitments has not been included in the ROIC denominator, downward pressure on GPU and computing pricing is already becoming apparent, and the return question for the even larger "Phase Three" capital expenditure 鈥?about $4.14 trillion from 2028 to 2030 鈥?has been explicitly left by Goldman Sachs for separate discussion.
Three phases of capital expenditure: scale rising step by step
Goldman Sachs divides this AI computing construction cycle into three phases: Phase One (2023 to 2025) total capital expenditure of about $633 billion, averaging about $211 billion annually, mainly used for training infrastructure buildout by a small number of foundation model companies; Phase Two (2026 to 2027) total capital expenditure of about $1.73 trillion, averaging about $863 billion annually, with the demand curve extending toward inference and computing supply remaining tight; Phase Three (2028 to 2030) total capital expenditure reaching about $4.14 trillion, averaging about $1.38 trillion annually, when capital intensity is expected to ease marginally and hyperscale cloud providers may gradually enter a "harvest mode."
It is worth noting that since the beginning of this year, consensus expectations for 2026 to 2027 capital expenditure among five listed hyperscale cloud providers (Alphabet, Microsoft, Amazon, Meta, and Oracle) have been revised up by about 66%, a combined increase of about $754 billion, rising from $1.14 trillion to $1.89 trillion. Goldman Sachs said its own forecast for 2027 capital expenditure remains above market consensus and believes this forecast is "closer to what buy-side investors actually expect."
From capital expenditure to revenue threshold: Goldman Sachs' calculation framework
Goldman Sachs' calculation logic is clear: within the $1.73 trillion of Phase Two capital expenditure, about 70% ($1.21 trillion) is computing investment (servers, chips, networking), and about 30% ($518 billion) is 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).
On key assumptions, Goldman Sachs uses a cost of about $42 billion per gigawatt of computing capacity, calculates computing assets over a 5-year depreciation period and infrastructure shells over 15 years, assumes annual operating costs of about $836 million per gigawatt, and assumes utilization rates of 60%, 80%, and 85% for 2028 to 2030, respectively. Under a 15% ROIC target, the six companies would need to generate about $129.5 billion in after-tax net operating profit (NOPAT) annually, corresponding to about $164 billion in annual earnings before interest and taxes (EBIT). Adding annual depreciation and amortization of about $276 billion on Phase Two assets, the final result is that annual revenue would need to reach about $465 billion to $476 billion, totaling about $1.42 trillion over three years, corresponding to an EBIT margin of about 35%.
Depreciation pressure: the core risk that is underestimated
In the above calculation, the most striking figure may not be the revenue target itself, but the scale of depreciation 鈥?annual depreciation and amortization on Phase Two assets alone reaches as much as $276 billion, about 1.7 times the EBIT these assets are required to generate. This means the economics of this AI investment cycle depend to a large extent on the actual useful life of GPUs. Goldman Sachs uses a 5-year depreciation period; if the actual economic life is only 3 years (some skeptical analysts and Nvidia's own annual product iteration pace both point in this direction), hyperscale cloud providers would be forced to continuously purchase new-generation chips at a higher frequency, and the revenue threshold would rise substantially.
Goldman Sachs' sensitivity analysis shows that within a range of ROIC targets from 0% to 30% and capital expenditure per gigawatt from about $34 billion to about $51 billion, the cumulative revenue required by the six companies from 2028 to 2030 would range from about $900 billion to $1.9 trillion, equivalent to annual revenue of about $6.2 billion to $18.6 billion per gigawatt. Even in a zero-return scenario, merely covering depreciation and operating costs would still require $920 billion in revenue.
Cloud backlog: the most important "safety cushion"
The core evidence Goldman Sachs provides for the feasibility of the revenue target is cloud backlog data. As of the second quarter of 2026, AWS, Azure, and Google Cloud together had a backlog of about $1.69 trillion, up about 1.5 times from the beginning of the year, with public cloud revenue growth accelerating to more than 45% year over year. Measuring only these three companies, their approximately $1.22 trillion in Phase Two capital expenditure corresponds to about $1 trillion in revenue demand for 2028 to 2030, equivalent to about 59% of the current backlog 鈥?and this calculation assumes zero growth in backlog thereafter. By company, the required revenue accounts for about 40% of Microsoft's backlog, 68% of Amazon's, and 78% of Alphabet's. Goldman Sachs characterizes this assumption as "possibly conservative."
However, the report also notes that the above backlog data includes non-cloud business commitments and is not strictly purchase orders; in addition, a considerable proportion of orders comes from a small number of AI labs whose ability to pay depends heavily on continued access to capital market financing, creating customer concentration risk.
The other side from Morgan Stanley: the financing gap is also enormous
Comparing Goldman Sachs' revenue threshold with Morgan Stanley's research provides a more complete picture of both sides of the issue. According to Morgan Stanley estimates, GenAI investment can achieve incremental ROIC of 25% to 50% 鈥?about 31% for GPU leasing, about 46% for model APIs based on proprietary infrastructure, and about 25% based on third-party infrastructure. Morgan Stanley assumes annual GPU leasing revenue of about $22.9 billion per gigawatt and model APIs of about $30.4 billion per gigawatt, both higher than the $11.6 billion per gigawatt corresponding to Goldman Sachs' 15% ROIC target. The conclusions of the two institutions are broadly consistent: if GPU leasing prices and computing pricing remain near current levels, the revenue targets are mathematically achievable.
However, another earlier Morgan Stanley study revealed pressure on the financing side: according to its estimates, global data center capital expenditure from 2025 to 2028 will be about $2.9 trillion, while hyperscale cloud providers' own cash flow can cover only about $1.4 trillion. The remaining financing gap of about $1.5 trillion would need to be filled through about $800 billion in private credit, about $200 billion in corporate bonds, about $150 billion in ABS/CMBS, and about $350 billion through other channels. Goldman Sachs' latest forecast shows that Phase Two and Phase Three combined capital expenditure alone reaches about $5.9 trillion, far exceeding the above estimate.
Blind spots in Goldman Sachs' framework: off-balance-sheet commitments and Phase Three
Goldman Sachs acknowledges in the report 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 about $3 trillion and are still growing rapidly. These commitments also need to generate returns or at least be repaid, and the credit market has already reacted to this.
In addition, Goldman Sachs' framework covers only Phase Two. If the same revenue-to-capital-expenditure ratio (about 0.82 times) is roughly applied to the Phase Three forecast of about $4.14 trillion, hyperscale cloud providers would need to generate an additional $3.4 trillion in revenue on top of the $1.42 trillion Phase Two revenue target. Peter Berezin of BCA Research went further, pointing out that if data center spending remains at current levels, as much as $10 trillion in annual AI revenue may be needed to monetize all capital expenditure. Goldman Sachs maintains a "buy" rating on all six companies and characterizes the recent compression in returns as "a natural result of a large-scale upfront investment cycle, rather than evidence that the economics of AI are structurally poor." But this statement holds equally true in both bull and bear narrative frameworks.
Comments