The bond market said something definitive this week: risk appetite tightened fast. The ICE BofA US High Yield option-adjusted spread widened to 3.47 percentage points on November 18 from 3.20 on November 12, rising every session. Some AI infrastructure and neocloud credit has been trading at distressed levels. But a spread widening is not the same thing as “credit markets stopped funding new AI deals entirely”, and it is not proof of an across-the-board primary-market shutdown.
The equity side of this trade has been brutal enough on its own. IREN was down 79% from its 2025 high, and CoreWeave sat 74% below its high from last year. On Friday, IREN fell nearly 16%, closing at $12.12 and hitting a new 52-week low of $12.06. CoreWeave posted a third-quarter net loss of $709 million. Net interest expense reached $846 million, and the company guided fourth-quarter interest expense to between $1.05 billion and $1.15 billion. Those are not speculative worries. Those are filed numbers.
The debt picture is worse. CoreWeave’s 9.75% bond due in 2031 has been yielding around the mid-teens, roughly 16% as of late this week. CoreWeave’s five-year credit default swaps have been trading around the low-1,000s in basis points in recent weeks, a level markets generally associate with distress and elevated default risk. The exact implied default probability depends on recovery assumptions and the pricing model, but the signal is the same: lenders are demanding crisis-level compensation to own the risk.
The investment committee question this weekend is whether credit has been right all along and equity is simply catching up. Bond managers get paid to care about downside first. When AI-linked high-yield paper prices materially wide of comparable BB and B-rated debt, that is a crowd of professional lenders saying they see asymmetric risk in this cohort specifically, not in junk broadly.
Analyst commentary has identified roughly $6 billion of remaining fiscal-year 2027 capital expenditure as a funding risk for IREN. Management has also flagged that capital markets have become more selective. That is a company acknowledging, in its own words, that the credit window has tightened around it.
Pimco offered the most credible counterargument this week. Pimco argued that spreads on bonds financing AI data centers may tighten relative to debt issued by the technology companies that use those facilities, if slower construction curbs new issuance. Still, the opportunity will not extend evenly across AI credit. Pimco has been more cautious on leveraged infrastructure providers whose interest expenses, leases, and chip-purchase commitments may persist even if construction setbacks delay the revenue needed to service them. That last sentence is the whole problem for CoreWeave and IREN in a single clause.
The divergence between credit and equity is not unique to AI infrastructure names. Corporate bond spreads across the broader technology and telecom sector have been widening while equity markets have yet to reprice the same risk.
Pimco wrote: “We expect a longer, and less linear, AI build-out rather than a materially smaller one.” That would spread financing needs over a longer period and ease near-term bond supply. A longer build-out benefits the hyperscaler-backed investment-grade bonds Pimco favors. It does not help neocloud operators who borrowed at 9.75% coupons and now watch those bonds yield around 16%.
The practical implication is a split credit market within AI itself. Hyperscaler-adjacent investment-grade paper may well tighten, as Pimco expects. The neocloud tier sits in a different universe.
Stocks to Watch
CoreWeave (CRWV) is the clearest stress test. The stock trades at 2.5 times trailing sales, down from 15.8 times at the end of September 2025, and gross margin fell to 8.5% over the four quarters to September 30, 2026, from 18.6% a year earlier. Revenue growth is real. The interest bill is growing faster.
IREN (IREN) enters this weekend at a 52-week low with a roughly $6 billion funding gap tied to planned FY27 capex and a high-yield market that has become far less welcoming to leveraged AI infrastructure stories. The stock has fallen far enough that two sell-side firms see 400%-plus upside. That gap between the equity and credit view is the entire debate.
Nebius (NBIS) fell on Friday as the market’s Anthropic-driven AI optimism faded; Nebius dropped 11.7% by mid-session, extending recent neocloud weakness. Its credit profile is cleaner than CoreWeave’s, but it trades in the same risk bucket in a risk-off market.
Oracle (ORCL) occupies a different tier. Oracle’s five-year CDS moved wider in November and has been cited around the low-to-mid 300s in basis points in recent weeks. Wide for investment-grade, but nowhere near neocloud distress levels. Oracle is what AI credit stress looks like when a balance sheet can absorb it.
HYG, the iShares iBoxx high-yield ETF, is the broader transmission mechanism. If AI-linked high-yield stress spreads into non-AI junk, the spread between the neocloud disaster and the broader market closes in an uncomfortable direction. That is the tail risk equity investors are not pricing yet.
