The Same AI Debt, Priced Two Different Ways

The bond market now charges Meta more to borrow than TSMC. The margin is thin, but the direction is new: borrowing costs for U.S. technology companies have converged with emerging-market debt indexes, to within a few basis points of each other — a basis point is one-hundredth of a percentage point — where a spread of 30 to 50 points used to be normal ^[1]. Memory maker SK Hynix pays just 9 basis points more than Amazon to raise money. The same stretch of days put the S&P 500 and the Nasdaq 100 at record highs. Last week's split on this beat ran between what institutions said and what they did ^[2]. This one runs between two prices. The credit market's account of the buildout is the slower and more arithmetic of the two. The cost of insurance against hyperscaler defaults — credit default swaps, contracts that pay out if a borrower stops paying — has widened roughly 60 basis points relative to equivalent insurance on banks since October 2025 ^[3]. Apollo's chief economist, Torsten Slok, describes what the bond market is repricing.

What the market is repricing is hyperscaler credit fundamentals, namely a debt-financed AI capex cycle with rising leverage, negative free cash flow and uncertain payback on depreciating assets. — Torsten Slok

Beneath that sits a plain inversion. Analysts now project that 2026 capital spending at Alphabet, Amazon, Meta, Microsoft and Oracle will exceed the five companies' combined operating cash flow for the first time ^[4]. At all four of the biggest spenders, infrastructure spending is running faster than revenue, draining reserves and turning free cash flow — the cash left after the buildout is paid for — negative ^[5]. Not everyone in the bond market reads that as a verdict. JPMorgan's Stephanie Aliaga has a specific reason for doubt, and it is a sizable one ^[6].

We think that the market is very capable of absorbing this new issuance. And if anything, it might just enable this AI boom to be sustainable. — Stephanie Aliaga

The equity market hears the same arithmetic and moves the other way. UBS's analysts look at those record highs and reach a verdict that is not cautious.

parts of the market like the Nasdaq, which are not yet hit highs, are actually looking outright cheap considering the scale of sales growth that they are, uh, they are generating. — Rickie Fowler

UBS's case rests on a number that barely existed a year ago: the cloud revenue backlog tied to data centers has grown to $2.4 trillion — equal to all existing cloud revenue — and Alphabet says it can convert about half of that within two years ^[7]. A summer of de-risking cut stock volatility from over 50% to the mid-30s, which let leveraged funds rebuild positions.

this is just a pretty unprecedented growth era. — Rickie Fowler

At the infrastructure layer the money is not a projection. Nvidia booked $96.2 billion in a single quarter, with data-center revenue up 117% to $89 billion ^[8]. Jensen Huang's frame for that number is maximalist.

The buildout of AI factories — the largest infrastructure expansion in human history — is accelerating at extraordinary speed. — Jensen Huang

CoreWeave, one of the builders renting that capacity back out, says its scale is beginning to show up in its own results.

CoreWeave reached an important inflection point this quarter as our scale began to translate into expanding operating leverage. — CoreWeave

Here is the stitch between the two markets, and it is short. Part of the demand the equity leg is counting is the credit leg's spending seen from the other side. Nvidia has committed $279 billion to memory purchases through 2032, and its suppliers have booked that back as guaranteed minimum revenue — roughly $100 billion at Micron across sixteen agreements, and at least $93.9 billion at SanDisk ^[8]. The same dollars surface on both sides of the split. At the far end of the chain, the labs that are supposed to generate the revenue are the unprofitable link. Anthropic lost $42 billion in a year while planning to spend $518 billion, and the sector needs somewhere between $300 billion and $1 trillion in annual revenue just to break even ^[9]. Arthur Hayes draws the line where the revenue stops.

If you study financial history and you study every single major technological rollout, it always is overbuilt. — Arthur Hayes

The only participant with no stake in which market is right is the one being paid to stand between them. The U.S. securities industry is pacing toward a record $90 billion in profits this year, and the New York comptroller attributes it directly to the financing demands of the AI boom ^[10].

It's getting close to as good as it gets. — Jamie Dimon

Global equity issuance crossed $1.08 trillion in nine months — only the second time that mark has been passed — with AI-linked deals nearly half of third-quarter activity ^[11]. The fee stream runs no matter which market turns out to be right. The official sector is already doing the downside arithmetic that no market price settles. Australia's central bank now models a scenario in which Australians hold roughly 5.4% of their financial wealth in AI stocks, and a permanent 20% drop in AI prices cuts household consumption by 0.7% — or 2.5% if the fall spreads to the rest of the market ^[12]. The question that arithmetic raises is live in Washington and unanswered: if credit is right and equities are wrong, does anyone catch the fall? Treasury Secretary Scott Bessent favors programs that support markets. The Fed chair, Kevin Warsh, wants them left to trade on their own ^[13]. The argument over the rescue is already underway, before any crash has arrived to make it real.