Goldman Sachs Forecasts Continued Growth in AI Capital Expenditure, Says Revenue Must Catch Up

Goldman Sachs strategists expect capital expenditure by U.S. hyperscale cloud companies to rise to roughly $1.2 trillion by 2027, but they believe external financing needs and supply bottlenecks will increasingly constrain the pace of that spending growth.

The Goldman Sachs team led by Ryan Hammond believes next year's capital expenditure will exceed the consensus estimate of $1.1 trillion and forecasts it will reach $1.4 trillion in 2028. However, they noted that "both the pace of capital expenditure growth and the magnitude of upside surprises will slow compared with recent quarters." The report mentioned that this year's capital expenditure is expected to double from 2025, reaching $800 billion.

Hammond pointed out that hyperscale cloud companies' capital expenditure has already exceeded the cash flow generated by their operating activities, meaning that achieving significant future growth will increasingly require raising funds through debt and equity issuance. According to the research report, these companies will need to generate about $300 billion in AI-related revenue each year in the coming years to break even on their 2026-2027 investments. For hyperscale cloud companies to achieve solid investment returns, consumer and enterprise spending on AI applications would need to reach about $1 trillion annually.

Although hyperscale cloud companies' current estimated AI revenue is not yet sufficient to cover capital expenditure and achieve break-even, their revenue is growing rapidly, and early signs of investment returns have already appeared. In the second quarter, cloud business growth at Amazon, Google, Microsoft and Oracle accelerated to 48%; meanwhile, combined cloud business revenue backlog at Amazon, Google and Microsoft reached $1.7 trillion.

The research report wrote that investor skepticism has already been reflected to some extent in stock prices, valuations of AI infrastructure have declined, and the price-to-earnings ratios of hyperscale cloud companies have fallen to their lowest levels in more than a decade.