Goldman Sachs: Hyperscaler Capex Could Reach $1.1 Trillion by 2027, AI Volatility Risks Rising
N.R. Finch
Goldman Sachs raised its 2027 hyperscaler capex forecast to $1.1 trillion, nearly 20% above consensus; but the bank warns that higher spending is driving higher volatility — and with U.S. equity earnings yields now below Treasury yields, the pricing logic underpinning stocks faces a fundamental challenge.
How far above consensus is Goldman's forecast?
Analyst consensus puts 2027 hyperscaler — Amazon, Microsoft, Google and peers that run massive data centers — capex at $920 billion, with growth slowing from 84% in 2026 to 22%.
Goldman calls that too conservative. Drawing an analogy to railway and auto-industry build cycles, the bank argues AI investment should reach 2–3% of GDP — implying roughly $1.1 trillion and ~45% growth.
In a more extreme upside scenario, factoring in hyperscalers' cash-flow generation and investment-grade credit capacity, capex could hit $1.4 trillion — growth of 89%.
This means → Goldman is not betting on whether AI investment continues; it is betting the scale will far exceed what the market expects.
The AI sector has rallied hard — what is driving it?
Goldman notes that AI infrastructure stocks have risen mainly on earnings growth, not valuation expansion. In plain terms = prices went up because these companies actually earned more, not just because sentiment ran hot.
But the report warns that valuations have recently begun to expand and positioning is increasingly crowded — signaling rising volatility ahead.
Hyperscalers have recently tapped equity markets for funding. This reflects their need for continued positive revenue revisions to support current share prices — if revenue expectations are revised down, valuations come under pressure.
Where does the money come from — and what does it cost?
In 2026 alone, hyperscaler AI capex is projected to consume roughly $770 billion in cash flow.
Return on equity (ROE — a measure of how much profit a company generates per dollar of shareholder capital) for large-cap tech is expected to drop about 7 percentage points, with sustained AI capex expansion as the primary drag.
This means → even as revenues grow, massive capital spending is eroding profitability. Every dollar of return now requires more capital behind it.
Equity yields below Treasuries — is that normal?
The S&P 500 earnings yield — company profits divided by share price — sits at roughly 3.6%, about 85 basis points below the 10-year U.S. Treasury yield.
In plain terms = textbooks say stocks are riskier than bonds, so their yield should be higher. That relationship has flipped — investors are taking more risk for less return.
Goldman offers two explanations: either bonds are mispriced (implying a sharp inflation drop, massive deleveraging, or recession ahead), or stocks are overpriced. With U.S. May CPI at 4.2%, Q1 GDP growth at 2%, and the fiscal deficit hitting a record $1.2 trillion in the first eight months of this fiscal year, the bank leans toward stocks being expensive.
How does this mismatch resolve?
Goldman labels the situation a "conundrum": the recent tech rally has been driven largely by AI enthusiasm, yet AI's actual contribution to earnings remains quite limited.
This reflects a core risk — the market is paying a premium for an earnings story that has not fully materialized.
Whether this structural mismatch can self-correct without triggering a sharp adjustment is, in Goldman's view, the key variable for testing current market valuations.
Content is for reference only, not financial advice.