Goldman Sachs Estimates Cloud Computing Companies’ Debt at $420 Billion in 2027
Key Facts
Goldman Sachs Research expects global debt issuance by hyperscale cloud computing companies to reach $420 billion in 2027, up from an estimate of $250 billion for fiscal 2026, Reuters reported on November 9. The estimate puts borrowing at the center of the debate over financing AI infrastructure. On the same day, Reuters reported that banks and credit investors are stepping up scrutiny of AI debt deals and hesitating to increase their exposure to borrowers. This comes as the market’s view of the sector’s growth shifts following a wave of stock selling. Goldman Sachs shares closed at $865.82 on November 6, according to EL7 data.
Goldman Sachs CEO David Solomon said the bank advises and finances companies working in AI, and that building the technology’s infrastructure requires a massive mobilization of capital. He described the eventual scale of investment as uncertain, saying: "I don't think anybody really knows whether we're going to spend $5 trillion building infrastructure or $10 trillion or more." This range reflects the breadth of uncertainty over the scale of spending required, even as computing, data centers and power continue to need financing. Solomon said he was not concerned about AI’s long-term sustainability, but described the investment path as nonlinear, with periods of acceleration and slowdown. He thus distinguished the long-term continuation of the technological shift from fluctuations in the pace of spending along the way.
Investors’ concerns also relate to the expected returns on spending and the cost of financing it. Solomon said higher capital costs change investors’ outlook, and that market volatility reflects questions about the pace of development, the most attractive opportunities and capital flows; higher interest rates also put pressure on some business models. He added that the market could see a recalibration as investors reassess the scale of computing demand, its growth trajectory and the infrastructure needed to support it. In a sign of the breadth of the retreat, Reuters said a stock-market drop on the Thursday before its report wiped $2.5 trillion from the Nasdaq 100’s market value and pushed it into correction territory. This divergence between large financing forecasts and credit caution makes it more important to test whether computing demand can justify the spending.
The financing cycle extends beyond cloud computing companies alone. Goldman Sachs said the expansion of AI-related investment is generating demand for cloud service providers, data center builders and energy businesses. One bank official described this reach as "a whole second generation of the AI economy." The bank also said technological development will drive recurring investment cycles for years, tying financing to renewed infrastructure needs. Goldman Sachs Research estimated that global power demand from data centers will rise 170% between 2025 and 2030, with the United States accounting for more than 60% of the projected increase. These figures highlight that expanding computing requires power supplies and facilities alongside financing for servers and data centers.
The power grid remains an additional execution constraint on the spending cycle. Goldman Sachs said the queue of generation and storage projects associated with U.S. grid connection requests stood at about 2,600 gigawatts, more than twice the grid’s current installed capacity, according to an article published by the bank on October 19, 2026. Alongside energy needs, the bank noted that capital markets activity does not depend on AI alone: alternative assets continue to grow, while financing needs for mergers and acquisitions are renewed. This places borrowing forecasts within a broader financing market in which multiple sectors compete for capital. Goldman Sachs warned that assumptions about AI monetization and the viability of expansion continue to change, adding: "There will be market corrections."
In a separate estimate, Goldman Sachs put total underlying AI-related capital expenditure at about $7.6 trillion between 2026 and 2031, spanning computing, data centers and energy. The figure provides a measure of the financing required across the infrastructure chain, but comes alongside the bank’s warnings that assumptions about demand and returns can fluctuate. Expected debt issuance in 2027 remains an indicator of one potential financing channel, as banks and investors continue to scrutinize deals involving AI-related borrowers. Solomon said a reassessment of computing demand could reset market expectations at some point. In his view, that reassessment will determine how closely planned spending aligns with actual demand for computing capacity.