Goldman Sachs strategists predict that the five largest American hyperscalers will allocate around $1.2 trillion for capital expenditures in AI infrastructure by 2027. This represents a 50-54% increase from the $800 billion estimate for 2026 and exceeds Wall Street's consensus forecast of approximately $1.1 trillion. In an optimistic scenario, the figure approaches $1.4 trillion. Over a longer term, Goldman estimates cumulative spending on AI infrastructure at $7.6 trillion from 2026 to 2031.
The study was led by strategist Ryan Hammond. The five companies driving this spending are Amazon, Alphabet, Microsoft, Oracle, and Meta. Each is accelerating the construction of computing capacities and data centers to train increasingly complex models and enable large-scale inference. Throughout 2026, Goldman’s team has repeatedly raised estimates, citing strong quarterly results from hyperscalers in the second and third quarters, which indicated previous forecasts were overly conservative.
Spending on AI infrastructure nearly doubled in 2026. Goldman expects a slowdown to 54% in 2027 and to 12% in 2028, when total spending could reach $2 trillion. However, there is an important caveat in the analysis: the five hyperscalers will need around $300 billion in annual revenue related to AI just to break even on their infrastructure investments.
Strategists highlight three key constraints that could slow or alter this wave of spending: energy supply, workforce availability, and memory chip shortages. Energy is the most fundamental constraint: training and operating large models require vast amounts of electricity, and data centers are already placing a strain on energy grids in key markets.
Related articles
