Goldman Sachs has analyzed the capital expenditure of six major hyperscalers: Alphabet, Microsoft, Amazon, Meta, Oracle, and SpaceX. The firm divides the infrastructure buildout into three phases, with capital spending totaling $633 billion in 2023–2025, projected at $1.73 trillion for 2026–2027, and reaching $4.14 trillion for 2028–2030.
Annual infrastructure spending is expected to hit $800 billion in 2026, rising to $1.2 trillion in 2027 and $1.4 trillion in 2028. These projections exceed current Wall Street estimates. As of Q2 2026, the combined contract backlogs for AWS, Azure, and Google Cloud reached $1.69 trillion, marking a 152% year-over-year increase.
To break even on infrastructure investments, these companies require $300 billion in annual AI revenue. Achieving satisfying returns for providers and developers would necessitate $1 trillion in annual revenue. Goldman estimates that to reach a 15% annualized return on capital, the group must generate $1.42 trillion in cumulative revenue between 2028 and 2030.
The firm notes that compute resource supply constraints are a primary driver of the current expenditure acceleration. While capital is being deployed immediately, the revenue required to justify these investments is not expected to materialize until the 2028–2030 period.
