CoreWeave is financing long-lived AI infrastructure while its computing assets can become obsolete quickly. This tracks a public default, distressed restructuring or rating fall to B-minus or lower.
40%likely
Thin sourcingNext check
WatchingHedge · pays if the others are wrong
Why it is different
This is the book’s credit hedge against the AI buildout: the outcome uses public filings and rating actions, not an inaccessible derivatives quote.
Why it matters
It tests whether contracted demand can outrun heavy financing needs, negative free cash flow and fast asset obsolescence.
Next check · 20 Nov 2026
Does CoreWeave’s Q2-2026 report show interest costs above 28% of quarterly revenue? (Q1-2026: 25.8%.)
How we scored it ·
from how similar cases went
If right, how big4/5How overlooked3/5Upside vs downside5/5Source trust3/5Independent sources3/5
The exact wording we score
On or before 31 Dec 2028, at least one CoreWeave-only event occurs: (a) a bankruptcy filing, missed payment, distressed exchange, or court/creditor restructuring; or (b) S&P's issuer credit rating falls by at least two notches from the 9 Apr 2026 B+ baseline to B− or lower. Resolves YES on the first qualifying event; NO if neither occurs by the deadline. No other issuer can be added after the call. Market spreads remain diagnostic but cannot resolve the forecast because no complete public CoreWeave CDS history was demonstrated at review.
PROOF
Evidence and provenance
The links below are the evidence recorded when this forecast was scored. Each note states source quality or bias; a citation is not an endorsement.
Recorded sources · 3 citations / 3 independent origins
2026-07-28 | sell-side: ratings and refinancing remain constructive; mainstream press: AI-bubble risk is crowded but mostly framed through equities; priced market: CoreWeave CDS compressed to about 452bp in June 2026; specialist: credit analysis highlights GPU residual-value and tenor mismatch | E3
Revision history6 entries
Probabilities and interpretations are never silently overwritten. This is the append-only record of what changed and why.
Checkpoint resolved NO; P 45→40. Q2-2026 (SEC exhibit, 2026-08-11): net interest expense $640m on revenue $2,575m = 24.9%, under the 28% bar and below Q1's 25.8% (P 55, Brier 0.30). Revenue is compounding faster than debt service (+112% YoY, adj. EBITDA $1,510m, backlog ~$104bn). Only five points cut: total debt is ~$34.9bn, up ~$10bn in a quarter, net loss $626m, capex $6.4bn — the mismatch grows while the ratio improves. A NO on timing, not on the thesis. New checkpoint: Q3-2026 ratio above 26%, P=40.
Domain review revision: P 62→45. The 900 bp CDS leg and its checkpoint were removed because the project had not demonstrated a complete public CoreWeave CDS history capable of independent adjudication. The public outcome set is now SEC-filed credit events or S&P B−/lower. The expression may still use institutional credit protection, but market access is no longer confused with public forecast resolution.
The open issuer set and soft “material event” language are removed. CoreWeave is the only eligible issuer; default/restructuring, the two-notch B+→B− boundary, and the 900 bp/40-day CDS test are exhaustive pre-committed legs. The 62% component arithmetic now matches that fixed rule.
(2nd pass) — P 48→62, and the correlation tag was wrong in SIGN. Two findings. (1) The claim is a three-way disjunction ('at least one of (a) default, (b) downgrade cascade, (c) sustained spread blowout') over 2.4 years with an open-ended entity set and no threshold on (c). Scored as a thesis rather than as its own resolution rule: P(a)~0.55, P(b)~0.30, P(c)~0.65 correlated ≈ 0.78 as written. Raised to 62 rather than 78 because leg (c) has now been given a real threshold that forces it to test the actual thesis — GPU-specific spread widening while HY OAS stays contained — rather than resolving YES on a generic credit risk-off that would say nothing about GPU residual values. (2) corrgroup ai-capex → ai-capex-credit with corrsign: -. This signal is ANTI-correlated with the physical-constraint legs: the same Microsoft earnings sentence that confirms grid-gear kills this, and vice versa. It was being discounted as a correlated fifth leg when it is in fact the book's only short-AI-capex position — portfolio insurance against four long signals, and worth more than its standalone score implies.
v2 rescore. P 55→48. The credit is currently re-rating TIGHTER, which this file did not reflect. CoreWeave's 5y CDS collapsed from 881bp (Dec-2025) to as low as 452bp (early Jun-2026); it issued $1.25bn @ 9.625% and EUR 2bn @ 8.500% due Jul-2032 on 2026-06-11 into meaningfully oversubscribed demand; S&P revised the outlook to positive on 2026-04-10. Verified anchors: Q1-2026 total debt $24.859bn ($7.547bn current), net interest expense $536m on revenue $2.078bn = 25.8%; DDTL 4.0 rated A3/A(low) at SOFR+225; CFR Ba3/B+. Sourcing risk logged: the BIS study and the CoreWeave figures currently reach this KB only through one aggregator (TechTimes, Trust 2) — replace with bis.org and the filings directly.
created; trust/sources trimmed vs research draft. TODO: locate primary BIS study and rating-agency reports to raise confidence.
Method challenge
Adversarial review record
Admitted after revision
Structured internal role review tied to this frozen claim. It is not independent human peer review. Independent specialist review not yet performed
credit or macro specialist
market-expression and risk reviewer
filing and accounting reviewer
forecasting-method reviewer
Strongest specialist challenge
S&P currently rates CoreWeave B+ with a positive outlook, citing contracted demand, scaling execution, and potential stronger credit metrics. The bear case must overcome take-or-pay visibility and sponsor or customer support, not merely point to large gross debt.
Outside view
The relevant outside view is transition and distress frequency for a named B+ leveraged infrastructure issuer over roughly two and a half years. That is materially below the former three-leg 62 percent estimate once the unverified CDS outcome is removed.
Causal rival
Revenue can compound faster than interest expense, customers can absorb contracted capacity, useful GPU lives can extend, and refinancing can move maturities outward. Those mechanisms can preserve credit even if spot GPU rental prices fall sharply.
Measurement risk
A complete public CoreWeave CDS closing history was not demonstrated, so the spread leg could not be reproduced by a public reviewer. The revised outcome uses SEC-filed credit events and S&P's public issuer rating only; broad spreads remain contextual.
Residual risk
The evidence set still leans on low-quality transmission sources for several financial claims, and rating agencies can lag distress. SEC filings and the named S&P action are auditable, but the economic thesis may deteriorate well before either binary leg resolves. Re-reviewed 2026-08-26 after Q2-2026 filings: net interest expense fell to 24.9% of revenue while total debt rose about $10bn in the quarter to roughly $34.9bn; the probability was cut five points because the timing leg weakened, not because the mismatch narrowed.
COLDTELL