Gold fell yesterday.
That is the sentence to sit with, because it breaks the reflex every investor carries in muscle memory. Oil topped $100. The 10-year hit its highest level since November 2023. Iran-war uncertainty was rippling through every market. And the one asset that is supposed to catch bids in exactly that environment — the crisis hedge, the fear trade — went down.
It went down because the market is no longer pricing a shock as a reason for the Fed to ease. It is pricing a shock as a reason for the Fed to hike. Next week.
Read that inversion carefully. The latest CPI print plus surging crude "bolstered the case for the Federal Reserve to hike interest rates," per Bloomberg — and gold, sensing tighter policy rather than looser, sold off. When the safe-haven reflex breaks, it tells you the regime has changed underneath you.
So the question every AI-stock holder is quietly asking: does the discount rate finally matter now?
Last issue, on August 28, we filed AI's debt dependence as a forward risk — a thread to watch, a leverage story building offstage. This week it stopped being forward. Oil above $100, the 10-year above 4.9%, a rate hike priced into next week: the shock is live and it is priced in.
Here is the strongest argument that says it still doesn't matter, and it deserves a fair hearing because it is genuinely mechanical, not sell-side hand-waving.
Bank of America's case runs like this. A stock is worth the present value of future cash flows; higher yields raise the discount rate and shrink that value; the assets that break first are the long-duration ones promising profits in 2032.
But the AI trade, BofA argues, isn't a long-duration trade at all. Per BofA, Nvidia is up roughly 24% over four months, Microsoft about 25% — and both are earning enormous sums right now on an infrastructure build-out the cloud providers are bankrolling today.
That is cash landing today, not a promise for 2032. When earnings arrive now, the discount rate matters less.
And the kicker: sector earnings are growing faster than share prices. A stock up 24% whose earnings rose more than 24% has gotten cheaper. The trade is de-rating as it grows — the opposite of a bubble dynamic. BofA's own bubble indicator shows no bubble; its stress index shows derivatives markets calm.
History seems to back the "high bar" read. In the late 1990s, long-bond yields rose more than 200 basis points and the Fed hiked over 100 while the Nasdaq soared. If AI stocks survived that, why fear a move to 4.9%?
Because — and this is where the argument quietly dies — the 1990s comparison flatters the present. Rates rose then against a booming-growth backdrop. That gave you an earnings offset: rates up, but growth up too.
This yield move is the wrong kind. The 10-year is climbing on oil-driven inflation fear and a fiscal picture the market no longer trusts — a term-premium and supply move, not a growth move. When rates rise on inflation and fiscal supply, no earnings offset waits to catch you.
BofA half-concedes this. Its own tell is buried in the reassurance: the 30-year is already above 5.1%, and the firm flags the risk of a "disorderly bond market tantrum" if upcoming Fed meetings don't deliver the tightening the market wants.
Sit with that. The comforting case requires the Fed to hike enough to keep the long end orderly. The reassurance is contingent on tightening — which is not the environment long-duration growth equity was built to love.
Now turn the whole frame, because Goldman's rates desk supplies the piece BofA's mechanism omits.
Mark Wilson, Jan Scheffel and George Cole open Goldman's Macro Call with the real question: is this another inflation scare, or a lasting change in the level of interest rates?
Their answer reframes AI entirely. The buildout, they argue, "has become an additional macro variable: a capital-hungry private-sector claimant on the same pool of savings governments are already draining."
The Fed can move the front end. But the real driver is duration supply, fiscal issuance, and demand for capital. And AI is now on the demand side of that equation — per the Goldman framing, not as a stock story, but as a borrower.
Look at the size of the claim. Wall Street reportedly estimates AI-related corporate debt issuance will hit $500 billion in 2026 — roughly 20% of all US corporate debt raised that year.
The Dot-Com bubble peaked with tech at 14% of total corporate bond issuance. On these figures — though the two are not a strict apples-to-apples comparison — the AI buildout looks set to be a larger share of the credit market than tech was at the height of the last mania — and it is doing this with debt, not fortress free cash flow.
That is the mechanism BofA's model cannot see, because BofA is looking at equity multiples and Goldman is looking at the savings pool. The AI trade isn't merely exposed to the rate regime. It is one of the forces manufacturing it.
Put the two claimants side by side. Governments are flooding the long end with issuance. AI is adding a half-trillion in long-dated corporate borrowing on top. Both draw on the same finite pool of savings. Unless fixed-income demand expands fast enough to absorb both, the clearing mechanism is higher real yields and a wider term premium.
Which means the AI trade is quietly bidding up the very discount rate that BofA insists can't hurt it. It is a snake eating its own duration.
The AI trade isn't merely exposed to the rate regime. It is one of the forces manufacturing it. A snake eating its own duration.
This is why treating the Fed as the variable that saves you is a category error. The Fed sets the front end. The front end is not what discounts a hyperscaler's 2030 cash flows or prices a data-center bond. Duration supply does — and the Fed is a passenger on that, not a driver.
So stop watching for a pivot. Watch for the switch.
BofA names it precisely, even as it argues the switch hasn't flipped: the trade holds only as long as earnings keep outrunning price. The moment price appreciation runs ahead of earnings, the AI trade stops being an earnings trade and reverts to a pure rate trade.
That is the day to fear — and it arrives from two directions at once. Either the mega-caps rerate faster than earnings can justify, or the long end spikes enough to force the disorderly repricing. The 30-year is already at 5.1%. It does not have to move much.
And here is the cruelest part of the setup. BofA's resilience is concentrated. The S&P 500 is up about 12% year to date, but most of that has come from a handful of mega-cap leaders while the broader market churns.
In a growth-led tape, concentration reads as a moat — the strong compounding while the weak fade. In a fiscal-supply, term-premium tape, concentration reads as a liability. When the switch flips, there is nowhere to hide inside the index, because the index is the trade.
The Nasdaq's roughly 1% drop was reportedly the worst of the majors yesterday. Small, so far. Absorbed, so far. The tape is still doing the BofA thing — churning, not breaking.
But the reassurance rests on the Fed keeping the long end orderly while the market prices a hike, oil sits above $100, and AI itself keeps borrowing against the same duration governments are draining.
The comfort was always conditional. Now the conditions are the variables — and the Fed doesn't control them.