Goldman Sachs strategist Ryan Hammond estimates hyperscalers need about $300 billion in annual AI revenue within a few years just to break even on investments. For solid returns, Goldman says AI users would need to spend about $1 trillion a year on applications.
Goldman forecasts the five largest U.S. hyperscalers' AI infrastructure spending will reach $1.2 trillion in 2027, above the $1.1 trillion Wall Street consensus and roughly $800 billion they are on pace to spend this year. Its model puts total hyperscaler capex at $1.4 trillion in 2028, with growth slowing to 12% after 54% in 2027. Cloud revenue was about $70 billion above its pre-AI trend on an annualized basis in Q2 2026, while announced group backlogs exceed $1.5 trillion.
Buybacks by the five largest hyperscalers fell 64% year over year in Q1 as they shifted cash toward infrastructure; Goldman credit strategists project debt will finance more than a third of their 2027 capex. Goldman does not call the AI trade a bubble, but says investor skepticism about durable infrastructure returns is already partly reflected in stock prices. It still expects S&P 500 buybacks overall to exceed $1 trillion this year, supported by other sectors.
