Goldman Sachs: AI Is a Capex Cycle, Upward Pressure on Interest Rates Hard to Fade
Claire Weston
Goldman's macro trading desk published its H2 roadmap, arguing that AI is a capital-expenditure cycle, not a software cycle — physically closer to building railroads than to the dot-com expansion. That means rates face sustained upward pressure, not the decline markets expect.
Why does Goldman compare AI to railroads, not software?
Software cycles scale with almost no capital, which pushes equilibrium rates down. AI requires data centers, power grids, and cooling infrastructure — all heavy physical assets.
This means → AI competes with every other sector for the same finite pool of capital, driving rates up rather than down.
In plain terms = the market has been pricing AI as "the next internet." Goldman says it looks more like "the next railroad boom" — one saves capital, the other devours it.
Where does the money come from?
Hyperscalers — Microsoft, Google, Amazon — keep raising their capex budgets. Funding pressure has already spilled from corporate balance sheets into public markets.
Credit issuance is large and still growing; equity financing channels have opened in parallel — companies are borrowing on both the debt and equity lines at once.
This means → the private sector now competes with governments for the same shrinking pool of duration buyers. Fiscal-supply pressure has a second leg, further capping how far mid- and long-end yields can rally.
What is the equity rotation telling us?
Capital is rotating from AI capex players (hyperscalers) toward AI beneficiaries (downstream application companies). Goldman calls this the clearest market signal right now.
This reflects a shift: even equities are starting to question whether the money being spent will earn an adequate return on invested capital — no longer just chasing the narrative.
The return question remains unresolved. Goldman advises putting weight on what is already certain — capital competition is intensifying, and that part requires no guesswork.
What about inflation and physical bottlenecks?
If the capex buildout materializes, power, grid capacity, skilled construction labor, and cooling facilities will face real-world shortages.
This means → these bottlenecks slow the path back to normal inflation — even as monetary policy tightens, supply-side friction keeps inflation elevated.
AI may eventually deliver productivity gains that bring inflation down, but Goldman says the burden of proof sits with AI — until that proof arrives, inflationary pressure is concrete.
Will the Fed hike?
The three-month average of nonfarm payroll gains fell from 188k to 111k. Unemployment edged down, but participation declined in step — the two effects offset. Wage growth showed no clear acceleration.
Fed Chair Warsh's remarks at Sintra were read by Goldman as mildly dovish — he described inflation risks as having diminished, and his AI comments focused on long-run supply-side effects rather than near-term capital costs.
Goldman's call: the Fed can hike, but it is an option, not a necessity. If it does, no more than two to three moves. Front-end pricing is already adequate; Goldman prefers fading rallies further out the curve to express the capital-cost-up view.
Content is for reference only, not financial advice.