Goldman Sachs: AI Capex Under Pressure, Bond Market Emerges as Beneficiary
nashnova research
Goldman's EMEA Delta-One trading head Rich Privorotsky flags a striking reflexivity — every dollar cut from AI capex is one less dollar of bond issuance, giving the 10-year Treasury yield marginal relief, but also meaning the bond market structurally 'wants' AI trades to stay weak.
Why is the AI trade suddenly under pressure?
Privorotsky lists five drivers: Lam Research beat expectations but failed to clear the market's implicit ceiling — Tokyo Electron fell nearly 7% in a single session. Tesla cut full-year capex guidance from over $250 billion to roughly $200 billion, diverging sharply from SpaceX's expansion signals.
Meta raised its capex floor to $135 billion, but This means → a single-quarter spend of $49–59 billion in Q4, an extremely aggressive pace.
Texas Instruments reported a broad analog-chip recovery with strong data-center demand, yet its stock barely moved. DeepSeek's open-source inference framework DeepInfer promises major efficiency gains on existing infrastructure, forcing the market to re-examine whether incremental compute capacity is truly needed.
Privorotsky's summary: "None of these feels paradigm-shifting, but together they're enough to rattle already fragile sentiment."
What is the link between AI and the bond market?
The core logic is straightforward: every dollar trimmed from AI capex expectations removes one dollar of bond issuance demand, giving the 10-year yield (Wednesday's close: 5.12%) marginal support.
In plain terms = when Big Tech plans to spend less, it needs to borrow less, bond supply pressure eases, and rates don't have to climb as far.
This reflects a reflexive structure — the very factor creating the equity problem is helping to solve the rates problem.
How large is the hyperscalers' debt footprint?
As of September 30, hyperscalers — Amazon, Microsoft, Meta and peers running mega-scale data centers — carried roughly $3.8 trillion in off-balance-sheet liabilities, up about $700 billion from June 30.
Add roughly $250 billion in investment-grade bonds, and the AI ecosystem's Q3 duration risk (DV01 — a measure of how sensitive bond prices are to rate moves) was about 125% of total U.S. Treasury issuance over the same period.
This means → AI's financing footprint is now large enough to move the supply-demand balance of the entire Treasury market.
What does Meta's financial position reveal?
Meta added roughly $76 billion in new off-balance-sheet commitments in Q3 (mainly leases), pushing total off-balance-sheet liabilities to $747 billion. Free cash flow turned negative.
The CFO warned that "significant investment" would raise the cost of services. Privorotsky notes Meta has joined Amazon in the negative-free-cash-flow camp.
In plain terms = Meta is no longer generating enough cash to cover its spending — not because the business is weak, but because AI investment is so large it swallows the cash flow whole.
What does this contradiction mean for investors?
The flip side is uncomfortable for equity bulls: if rates can only ease when AI stocks fall, the bond market structurally "wants" the AI trade to stay soft.
Semiconductors fell in 10 of 11 sessions before Wednesday's bounce, and Goldman's U.S. desk noted that rally itself "looked like it lacked staying power."
This means → with the Fed still in a tightening cycle, the seesaw between AI equities and bonds is unlikely to resolve soon — when one side rises, the other comes under pressure.
市场有风险，内容仅供研究参考，不构成投资建议。