The forecast, from strategists led by Ryan Hammond, is more than 50 percent above the roughly $800 billion the bank expects the group to spend this year, and sits above Wall Street's consensus estimate of $1.1 trillion. The companies at the centre of the buildout are Amazon, Alphabet, Microsoft, Oracle and Meta.
The biggest investment cycle since railroads
Hammond's team frames the 2027 figure in stark historical terms: measured against GDP, it would be the largest investment cycle since railroad construction in the late 19th century. Back in June, Goldman had already warned that consensus estimates were far too low.
The trajectory tells the story. The group spent about $405 billion on capital expenditure in 2025. Goldman expects that to nearly double to around $750 billion this year, then jump to $1.2 trillion in 2027.
Growth then slows: the bank projects a 54 percent increase in 2027 followed by 12 percent in 2028, when spending would reach roughly $1.4 trillion. Goldman also estimates cumulative AI infrastructure spending could reach $7.6 trillion from 2026 through 2031.
The $300 billion break-even question
The uncomfortable part of the math is what the spending needs to earn back. Goldman estimates the hyperscalers need about $300 billion a year in AI revenue just to stop losing money on their infrastructure investments. Current results are still well short of that figure.
There are encouraging signs. Cloud revenue growth across the group jumped from 25 percent in 2024 to 48 percent in the second quarter of 2026, and hyperscaler cloud revenues are running about $70 billion above the pre-AI trend line.
But Goldman notes it is still unclear whether revenue growth at AI labs like OpenAI and Anthropic — which sit at the heart of the expectations backing the buildout — is fast enough to justify these investments.
What could slow the buildout
Spending is now rising faster than the companies' cash generation. Goldman says hyperscaler capital expenditure already exceeds cash flow from operations, forcing the group to lean more heavily on debt financing in the years ahead.
Physical constraints could also bite. The strategists flag shortages of power, skilled labour and memory chips, along with restrictions on data centre construction, as factors that could slow the pace.
Goldman also says the rate of upside surprises to consensus estimates will diminish relative to recent quarters, even as the absolute numbers keep climbing.
