The AI correction is now a ladder of borrowing costs

On October 17, two weeks before anyone voted, the flagship neocloud's bonds crossed the number the rest of the market has spent a month catching up to. CoreWeave's corporate debt — the paper of the best-known name in the new class of companies that rent out Nvidia's chips by the cluster — blew past a 15 percent yield while the campaigns were still busy arguing about something else ^[1]. A yield that high is a lender's way of saying the loan is being priced as a risk, not an ordinary bond. Critics were already drawing the comparison to Lucent Technologies in 2000, when a vendor's own financing kept a customer aloft ^[1]. The rout that has since been called the election's work began there, in credit, before a ballot was cast. It had reached the big index by November 3. The Nasdaq-100 peaked on October 6 and was in correction — down more than ten percent from its top — before the polls opened ^[2]. Goldman Sachs had already advised clients to overweight cash, to hold more of it than usual. Its chief strategist, Christian Mueller-Glissmann, gave the reason without mentioning politics.

AI risk has dominated cyclical risk. — Christian Mueller-glissmann

So what did the fourth of November add? A second constraint, laid on top of the financial one. The backlash against data centers is a year old and it crosses party lines: more than a hundred local moratoriums had been proposed by June, 71 percent of Americans say they don't want a facility near their home, and PJM, the grid operator for 67 million people, has proposed cutting data centers' power first in an emergency ^[3][4][5]. A Maine legislator put the geography plainly.

Red states, blue states, and purple states are having this conversation. — Melanie Sachs

What the election changed is that the backlash acquired a governor. In Texas, the state Greg Abbott promised would be the "epicenter of AI development," the winner has pledged an immediate moratorium on new data centers ^[6].

Texas will be the "epicenter of AI development." — Greg Abbott

For lenders, this matters because a moratorium map is a collateral map. The loan behind a neocloud's expansion is ultimately secured by an unfinished building — the collateral, the asset a lender can seize if the borrower stops paying — and a building is the one thing a new governor can stop. The giants have been buying their way around that risk for months, with Microsoft and Meta chasing nuclear deals and paying premium rates for power — adaptations priced for balance sheets the middle tier doesn't have ^[5]. Citrini Research, declaring the end of the trade, counted the Democratic victory among its reasons, citing the increased physical risk to data-center construction.

There are limits to how much investment hyperscalers can tap from capital markets. The trend has to be for deceleration — Citrinitas Capital Management Inc.

Now the ladder, one borrower per beat. Consider the calendar of November 3. That morning Goldman was telling clients to hold cash; the same day Alphabet was selling euro-denominated bonds — its second such sale of the year — because borrowing in Europe costs less ^[7]. Microsoft announced $25 billion in new corporate debt, and Meta plans to raise $95 billion by selling stock ^[8]. The giants buy money at wholesale. One rung down, the price of money is different. CoreWeave's bonds still trade near 15 percent, and HSBC's Abhishek Shukla has done the arithmetic: a projected $39 billion a year in capital spending against revenue guidance of $12 to $13 billion, a return on invested capital below the cost of borrowing ^[9][10].

Debt market woes along with moderating [large language model] revenues are exacerbating structural concerns for neoclouds. — Abhishek Shukla

Short-seller Vincent Daniel of Seawolf Capital puts the bet even more bluntly.

To own this equity, an investor must underwrite that CoreWeave can raise tens of billions of dollars in incremental capital on accretive terms—not a bet we are willing to take. — Vincent Daniel

Nebius is the same story told through a convertible — a bond that can be swapped for shares. That funding only worked while the share price stayed high, which is why revenue growing 454 percent a year hasn't stopped the stock from falling 40 percent from its June peak ^[11]. Growth was never the problem. What the share price funds is. At the bottom sit the bitcoin miners that pivoted to AI. They are selling record reserves of bitcoin because no one will lend to them on conventional terms, and the fund that tracks them is down 31 percent on the year ^[12]. Elon Musk warns of "guppy-level" consolidation among neoclouds without substantial revenue outside AI ^[12]. The obvious objection writes itself: the winners borrow plenty too. Three of the four hyperscalers are free-cash-flow-negative — their operations burn more cash than they throw off — after a record quarter of $201.9 billion in capital spending ^[8]. So the divide is not between the indebted and the clean. It is the coupon — the interest rate each borrower pays — and what stands behind it. Alphabet borrows in euros at tight spreads, thin margins over the benchmark rate, while CoreWeave pays 15 percent; Meta sells $95 billion of stock without collapsing its price while Nebius's convertibles are hostage to its own. Goldman's Gabriela Borges still rates CoreWeave a buy, arguing its delayed-draw loans are asset-backed project financing rather than plain corporate debt ^[10] — which is exactly the point, that the money now comes with a building attached. Nebius still sits on roughly $8 billion in cash against $8.5 billion of debt, up 117 percent on the year, and Amazon's Andy Jassy says AWS capacity is nearly sold out through 2027 ^[13][9]. Demand was never what broke. BlackRock's Rick Ryder reports strong appetite for big-tech debt; the appetite gathers where the paper is cheapest ^[14]. Jensen Huang, whose Nvidia has guaranteed billions to keep neoclouds building, named the constraint from inside ^[15].

If you look at the fundamentals, the demand for the neoclouds is off the charts. — Jensen Huang

Because there's just so much demand across so many sectors of AI, there's a scarcity of capital. — Jensen Huang

The week's tape shows the ladder being walked, not argued. JPMorgan's desk reported strong demand for Amazon, Microsoft, and Alphabet while CoreWeave and Nebius weakened, with the Dow outperforming the Nasdaq ^[16]. Citrini, even while declaring the end of the trade, pointed its clients not at the giants but at energy, financials, and consumer names ^[17]. The dip-buyers skip the middle too: David Katz's eight picks run from chip-equipment and power names to the hyperscalers, and omit the neoclouds entirely ^[18]. Across the Atlantic the same split showed up in a single market, the STOXX 600 touching near-record highs on steel and airlines the same day ASML fell on fear of U.S. regulation ^[19]. Fifteen percent is the price of the one asset a neocloud can pledge: an unfinished building in a state that may no longer want it built. In Texas, the governor-elect says she will ^[6]. The bond market set that number two weeks before the vote, and it has kept to its own calendar ever since. It never waited for the polls. It priced the building.