Goldman Sachs Estimates AI Compute Power Gamble: Six Tech Giants Need to Earn an Extra $1.42 Trillion from 2028 to 2030 to Sustain 15% ROIC

According to a report from Zhitong Finance APP, Goldman Sachs recently released a research report on the US technology sector, aiming to estimate the scale of AI economy required to justify the ROIC of hyperscale cloud providers' AI capital expenditures, focusing on a core market debate: as top US hyperscale cloud providers make massive investments in AI computing power, can these significant capital expenditures achieve reasonable returns in the future? The report establishes a quantitative assessment framework and conducts a return stress test on the 2026-2027 computing power investments of Google, Amazon, Microsoft, Meta, Oracle, and SpaceX, providing a reference benchmark for the profitability outlook of AI infrastructure investments.

Goldman Sachs believes that, faced with an explosive growth in large model Token consumption driving up computing power demand, US hyperscale cloud providers have seen dramatic changes in capital intensity to adapt to industrial transformation at the computational level. This round of significant investments on the part of enterprises is driven partly by real, current demand signals (with the market currently experiencing supply-demand imbalance), and partly to meet the computing power requirements of core clients for years to come.

Looking at the market performance this year, especially in light of the recently concluded Q2 earnings season, the market’s understanding of this theme has notably deepened. Judging from financial data and management commentary, investors are reaching two general consensuses:

(a) Hyperscale cloud providers can rely on previous (2023-2025) capital expenditures to achieve substantial, possibly above-expected, returns on investment through incremental revenue and operating cash flow;

(b) Driven by the tight supply of computing power and sustained acceleration of end-user demand, cloud providers are entering a new cycle of capital investment (2026-2027), with current pricing of computing power services significantly higher than the long-term benchmark price assumed by Goldman Sachs.

As a result, the focus of market debate in recent months has shifted to more long-term questions: for this round of capital investment during 2026-2027, how much return on investment can be generated in 2028-2030? What reference can be drawn from the ROIC realized by previous capital investment?

Goldman Sachs built an analysis framework in the report to estimate the incremental size of the AI economy required for US hyperscale cloud providers to achieve a benchmark return on invested capital (ROIC) at the current level of capital expenditures. The core of the analysis is to estimate the revenue threshold required for top US hyperscale cloud providers’ second-phase (2026-2027) AI computing power investment to achieve a 15% annualized ROIC.

Based on a series of assumptions: each gigawatt (GW) of computing power requires an average upfront investment of approximately $42 billion; 70% of capital expenditures go to computing hardware, 30% to data center shells; adopting a conservative depreciation policy and factoring in ongoing operating cost assumptions, etc. The results show that the six major US hyperscale cloud providers (Alphabet, Amazon, Microsoft, Meta, Oracle, SpaceX) need to generate about $1.42 trillion in cumulative revenue during 2028-2030 (corresponding to annual revenue of $11.6 billion per gigawatt of computing power) to cross the 15% ROIC threshold.

Although a significant expansion of short-term capital expenditure will suppress immediate returns (causing short-term performance pressure), Goldman Sachs believes that this round of capital expenditure has an attractive ROIC level, which will gradually be realized in medium- to long-term operating statements. In other words, short-term return pressure is a natural result of a large scale upfront investment cycle and does not indicate a structural flaw in the AI business model itself.

In Goldman Sachs’ previous reports on the Token economy, AI landscape on the consumer side, and the enterprise AI sector, it has been explained that: as market share changes and Token prices decline, AI penetration rate will instead increase, broadening application scenarios for computing power across both consumer and enterprise ends. The current market focus is on the evolution of leading foundational large model players—including market share and the pricing power of computing resources compared to open-source models. However, in Goldman Sachs’ view, the cost-performance pattern of computing power truly determines the overall scale and growth boundary of the AI economy.

There are various types of agents in the market with different models and capabilities, expanding along the cost-performance curve, unlocking a wide range of application scenarios from "commodity-level affordable intelligence (subject to greater price deflation pressure)" to frontier, cutting-edge intelligence (with stronger pricing resilience). Overall, this will support current capital investments by computing infrastructure providers.

In short, not every type of Token has the same commercial value, nor does every capital expenditure yield uniform returns. But on the whole, considering the vast market potential over the next 3-5 years, Goldman Sachs still believes that capital investments made in the next 18 months can generally achieve a good level of return.

This was also confirmed at the Goldman Sachs Communacopia Technology Conference, where participating enterprises conveyed three signals:

(a) The industry has shifted from the AI trial and exploration stage to the implementation stage; (b) Returns from efficiency gains and the product iteration cycle are accelerating; (c) Companies have shifted from purely pursuing the scale of Token consumption, to optimizing Token input-output ratios to enhance their own return levels, indicating that corporate adoption of AI will rise further (enterprise-level AI popularization will directly boost cloud provider revenue and incremental operating profit generated by earlier capital investments).

In addition, several AI agent products targeting consumers have recently been officially launched (most notably Meta’s Muse), signaling a paradigm shift in consumer AI: from conversational interaction products to agent-driven action execution products. The expansion of such agent platforms (with major tech firms highly likely to follow suit in the consumer AI platform layer) will drive an increase in medium- to long-term computing power demand. In the future, supporting monetization models will also gradually take shape, including subscriptions, advertising, e-commerce, and other paths. The boundaries of these business models are themselves continually merging and will become core drivers of future revenue growth.