Goldman Sachs says AI capex and government borrowing are pushing the cost of capital higher

Goldman Sachs has put a name to the force that is making borrowing more expensive for almost everyone: two large pools of demand are now competing for the same global supply of capital, and that competition is lifting the price of money.

The two pools are private and public. The first is the AI buildout. Capital expenditure by AA-rated issuers — companies with high-quality investment-grade credit ratings — rose 65% year-on-year in the second quarter, the tenth straight quarter of aggregate growth above 35%, according to a Goldman Sachs research note republished by ANI via EuropeSun. The second is government borrowing, which has resumed as the US, Europe, and Japan fund infrastructure, energy security, and defense.

"Cost of capital" is finance jargon for the price of borrowing money. When more borrowers compete for the same dollars, that price rises. Goldman frames the 30-year German and Japanese government bond yields as a marker: both were close to zero as recently as 2022 and have since repriced higher as AI-related and government demand grew. Long-dated government bonds are the baseline that other borrowing costs are measured against; their move higher lifts everything priced off them.

The bond market shows the same pattern on the corporate side. US convertible bond issuance — bonds that can convert into equity, popular with fast-growing tech borrowers because they let companies borrow at lower interest rates than straight debt — reached $135 billion year-to-date, with AI-related borrowers accounting for about 44% of the total, the note says. Goldman raised its full-year US investment-grade gross issuance forecast by $200 billion to $2.3 trillion, and estimates AI-related issuers at roughly a quarter of US investment-grade gross supply this year.

Goldman Sachs' companion piece "Tracking Trillions" lays out the assumption set behind the AI capex numbers: how much compute each model needs, what a gigawatt of data center costs, and where the spending lands on the income statement. The two notes share a single thesis: the AI buildout is large, durable, and funded by debt and equity at a scale that affects the entire cost of capital.

For general readers, the practical question is what this means for stocks. Goldman's credit and equity teams warn that the regime carries an equity risk: if corporate profit growth slows while borrowing costs stay high, equity prices face real downward pressure. The mechanism is that higher discount rates compress the present value of future earnings, and AI capex absorbs free cash flow that would otherwise return to shareholders.

Technology valuations have already moderated, and now sit below their 20-year median globally. The note treats that moderation as a partial market response, not the end of it. The warning lands as the constructive critique inside the source: the cost of capital can rise while AI capex absorbs free cash flow, leaving equity holders exposed if profit growth slows.

The falsifier is specific. The two-pool pressure eases if any of three things happens: AI capex contracts because model training efficiency improves faster than expected, government borrowing moderates as deficits narrow, or private credit expands enough to fill the gap. Until one of those breaks, the cost of capital is the price the system charges for funding both the AI buildout and the public balance sheet at the same time.

For now, the watch item is the next quarter of AA-rated capex. Ten consecutive quarters of more than 35% aggregate growth is a long run. A slowdown would be the first concrete signal that the supply side is starting to absorb the demand.