Goldman Sachs warns AI boom is running into a cost-of-capital problem

Big Tech has record profits and a spending habit that may be outgrowing them. Goldman Sachs says the AI infrastructure race is now big enough to push up the cost of money itself.

In research published in mid-to-late September 2026, the bank argues that hyperscaler capital spending is adding to global demand for capital at the same moment governments need to borrow more.

The spending math

The bank expects the top five US hyperscalers, Amazon, Alphabet, Microsoft, Oracle and Meta, to put approximately $800 billion into AI infrastructure in 2026.

That is nearly double the $413 billion they spent in 2025.

On the revenue side, Goldman projects AI-related cloud revenue of about $70 billion above pre-AI trends. The bank estimates roughly $300 billion would be needed to break even on the investment.

Goldman frames the resulting annual shortfall at about $230 billion.

Reports authored by Goldman strategists Peter Oppenheimer and Ryan Hammond link the rapid ramp in AI capex to weaker free cash flow across the technology sector.

Wall Street becomes the lender

The convertible bond market shows how fast this is happening. US convertible issuance reached $135 billion as of mid-September 2026, and AI-related borrowers accounted for 44% of it.

Investment-grade debt tells a similar story. Goldman’s credit analysts raised their forecast for US investment-grade issuance by $200 billion. They now see it potentially reaching a record $2.3 trillion in 2026.

Bullish on AI, cautious on the debt

Goldman Sachs Asset Management is underweight hyperscaler debt even while keeping a positive long-term view on AI.

The Brookings Institution estimates US AI infrastructure investment could total roughly $10.3 trillion by 2032, with debt playing a growing role in funding it.

What this means for investors and markets

The most immediate pressure point is valuation. Tech stocks priced on the assumption that AI revenue will eventually justify the spending face a test if the gap Goldman describes persists. A $230 billion annual shortfall is the kind of number that invites investors to revisit their models.

Credit investors have their own calculus. With AI borrowers making up 44% of convertible issuance and helping drive investment-grade supply toward a possible $2.3 trillion record, concentration risk is building in portfolios that may not have been designed to hold so much exposure to a single theme. GSAM’s underweight stance suggests at least some large investors want to be paid more for that exposure.

The things to watch are straightforward: whether AI cloud revenue starts closing the gap with capex, how much of the next wave of spending is funded with debt, and whether bond investors begin demanding wider spreads from hyperscaler issuers.