Goldman Sachs: AI Capex Pushes Long-End Rates Higher, Fed Now a "Passenger"
nashnova research
Goldman's cross-asset team argues that $500 billion in AI infrastructure financing is structurally driving global long-end rates higher, reducing the Fed's role from 'driver' to 'passenger' in this cycle.
How much are AI companies borrowing, and why does it move rates?
By early August, AI-related debt issuance had reached nearly $500 billion — roughly one-fifth of total U.S. investment-grade bond supply for 2026.
This means → hyperscalers — Amazon, Microsoft, Google and peers running massive server fleets — are now competing head-to-head with U.S. Treasuries for the same pool of bond buyers.
In plain terms = there is only so much money chasing fixed-income assets. When tech companies absorb a fifth of the supply, Treasuries and mortgage bonds must offer higher yields to attract capital — and that pushes long-end rates up.
Why is the Fed no longer the 'driver'?
Goldman's head of equity franchise sales, Mark Wilson, put it bluntly: "The Fed is more of a passenger in this discussion than a driver."
He identified five forces lifting yields: high fiscal deficits, hawkish central-bank rhetoric, commodity inflation, European political risk, and massive AI infrastructure borrowing — four of the five are beyond the Fed's or Treasury's easy repair.
This reflects a structural shift: pricing power over long-end rates is migrating from central-bank policy to real-economy capital demand. Whether the Fed cuts or holds matters less at the long end than it used to.
What does 'the end of the savings glut' actually mean?
Goldman's team believes the global savings glut is over. AI data centers, power grids, defense, manufacturing reshoring, and governments rolling over interest bills are all competing for the same finite pool of savings.
This means → the two-decade regime of "too much money chasing too few good assets" is reversing — now there is too much capital demand and not enough savings to go around.
The direct consequence: credit spreads, real yields, and term premium — the extra return investors demand for holding longer-dated bonds — are all rising in tandem.
Can AI's deflationary payoff offset this inflationary pressure?
Goldman's co-head of global short-term macro trading, Jan Scheffel, was explicit: "The AI investment boom is inherently inflationary."
Markets expect AI productivity gains to be deflationary, but Scheffel argues the economy is demanding the payoff before productivity actually arrives — the spending comes first, the efficiency gains lag behind.
In plain terms = costs rise now, savings come later. In the gap between the two, rates face net upward pressure.
$1.3 trillion in capex next year — where does the money come from?
Goldman flags that no one has yet mapped out how next year's $1.3 trillion in capital spending will be financed.
By then, hyperscalers' bond issuance will generate more DV01 — a measure of how sensitive a bond portfolio's value is to interest-rate moves — than the U.S. Treasury itself.
This means → tech companies' debt footprint will carry more weight in the rates market than the U.S. government's — an unprecedented situation, and the central unresolved question for the next phase of bond-market pricing.
市场有风险，内容仅供研究参考，不构成投资建议。