The wrong screen
This week the tape handed everyone a celebrity-macro headline: frontier commercial models cleared FedRAMP High and DoD IL-4/5, and a $25 billion debut bond settled for one of the highest-profile issuers in the market. Equity desks read both through the lens they've used all quarter — the AI cash-flow question, whether the buildout ever earns its return. That's the wrong screen for a bond desk. The signal that matters isn't on the equity tape; it's on the supply calendar and in the term-premium series. Both of this week's events point the same direction, and it isn't toward the front end.
Two supply waves, one stretch of curve
Start with scale. The five largest hyperscalers issued roughly $121 billion of US investment-grade paper in 2025 — more than four times their 2020–24 pace near $28 billion. Street estimates put 2026 net hyperscaler supply in the $130–150 billion range and AI-related IG supply broadly near $300 billion — call it $360 billion in 10-year duration equivalents once you account for maturity. Total US IG gross issuance is tracking a record, somewhere in the $1.8–2.25 trillion zone.
The defining feature isn't the headline number; it's the tenor. Data centers, power, and GPUs are multi-decade assets, so the paper funding them prints long. The roll-call says it plainly: a record $30 billion non-acquisition IG deal last October; $25 billion across eight tranches in February; a multi-currency program that placed 32-year and 100-year maturities; and this week's $25 billion debut carrying 2046 and 2056 tranches into a 3.5x book. This is duration manufactured at industrial scale and aimed squarely at the long end.
Now layer the second wave. Treasury is funding a structurally large deficit, the refunding mix has tilted toward coupons, and the marginal buyer base for duration is thinning at exactly the wrong moment. Foreign indirect-bidder participation at the 10- and 30-year auctions has plateaued. The Fed's balance sheet is no longer shrinking, but its stated intent is to rotate SOMA toward the front — aligning with Treasury's issuance mix and pulling official support out of the 10-year-and-beyond bucket. So the long end is being asked to absorb two simultaneous, structural, long-dated supply waves — sovereign and corporate — while two of its historically price-insensitive buyers step back. Both waves crest in the same part of the curve.
The regime that used to absorb this is gone
None of this would bite the way it now does in the old regime. For most of the post-GFC era the term premium was negative or trivial, and forward guidance did the work of anchoring the long end to the expected path of short rates. Cuts get priced, the 10-year is mechanically pulled lower, supply clears because duration is tethered to policy. That mechanism has broken in two places at once. First, Warsh's dismantling of forward guidance removed the anchor — the long end is no longer tied to a communicated rate path, because there isn't one. Second, and more important, the term premium has flipped from negative to positive and become a first-order driver of the long end rather than a residual.
This is where the lazy framing needs correcting. The line in circulation all year — "the bond market doesn't believe the Fed" — is wrong, and a desk should say so. Decompose the 10-year with the New York Fed's ACM model and the expected-short-rate component tracks the dots reasonably well. The market believes the path. What's repricing is the compensation for owning duration into a decade of fiscal-plus-capex supply and sticky inflation. The Fed can cut and the term premium can rise at the same time, leaving the long yield roughly where it sat — exactly the behavior the tape has shown. The 30-year printed a 19-year high near 5.2% in May; this morning it sold off into a broad risk-off session while the front end rallied. That twist steepener isn't noise. It's the regime: strip the anchor, flip the term premium positive, and the long end clears its own supply.
This is not (yet) a credit story
Be precise about what this is and isn't. Spreads are tight — IG broad OAS near 75, high yield near 276, both calm even on red-screen sessions. The issuers driving the supply carry pristine balance sheets, post-issuance leverage often in the 0.4–0.7x range against an IG average closer to 3x. The duration-absorption problem is a rates and term-premium problem, not a default problem. The credit tail exists and is worth naming: the buildout's anchor customers are pre-profit, venture-funded AI labs, and the revenue meant to service decades of data-center paper is contracted against names that have to monetize first. If that monetization slips, you get an obsolescence-and-collection story that would re-rate the sector. But that's the tail, not the base case — and the early warning won't be a spread blowout. It will show up first in new-issue concessions, which have run near 12 basis points against a market closer to 2.5 while deals still print 4x oversubscribed. Watch the concession, not the rating.
One nuance the on-balance-sheet figures miss: not every hyperscaler funds in public IG. At least one of the largest is routing the bulk of its buildout leverage through off-balance-sheet, fund-level structures, where the debt sits outside the corporate entity. The visible IG supply understates the sector's true financing footprint — which means the duration story is, if anything, larger than the league tables show.
Why "good news for AI" is more long-end supply
This closes the loop on the week. The IL-5 validation reads as a positive for the AI trade because it removes a demand-side tail — it lets commercial frontier models into defense and classified-adjacent workloads, a durable, non-cyclical buyer that doesn't flinch at a cash-flow-divorce headline. De-risk the demand side of the buildout and you extend its runway; extend the runway and you extend the corporate supply wave funding it. So the same datapoint the equity desk files under "AI bull case" lands on a bond desk as "the long-end supply calendar just got longer." The AI cash-flow panic on the equity tape and the term-premium repricing on the rates tape are not two stories. They are the same capex super-cycle seen through two screens.
The desk read
Five things to watch, in order of signal quality. Quarterly refunding statements — the maturity mix tells you how much long-dated sovereign paper is coming over the next three months; a tilt longer intensifies the pressure. Indirect bidder share at the 10- and 30-year auctions — the cleanest real-time read on whether foreign duration demand is stabilizing or still fading. The ACM term-premium series, updated daily — if it keeps climbing, the regime shift is durable; if it drifts back toward zero, the old model reasserts and the long end re-anchors. Hyperscaler capex guidance on the next earnings cycle — every upward revision is a forward-supply signal before a single deal prices. And the new-issue concession on the next mega-deal — the first place stress shows up if duration appetite finally thins.
The through-line fits on a card. The Fed handed the long end back to the market, and the market is repricing it for a supply decade. Front-end views still belong to the policy path; the long end now belongs to the term premium. Trade them as two different instruments — because that's what they've become.
Sources: Koyfin (rates levels), AM NY session. Supply and term-premium figures from Street and sell-side estimates, US Treasury refunding and NY Fed ACM data, and issuer filings; estimates are ranges, not point forecasts. For informational purposes only; not investment advice. Rich Petruzzo, CFA.