Bondholders reprice default risk before shareholders reprice growth: credit has no upside participation, so it reacts first when debt-funded expansion stops penciling. The 07/2026 specimen is AI capex. The GS hyperscaler bond basket’s spread went ~114 → 162bp in six weeks (“credit investors refuse to fund memory chip purchases any longer” per zerohedge) “while stocks completely ignore what is going on with bonds.”
The CDS dispersion is the real information: CoreWeave 701bp, Oracle 205, SpaceX 168 — the debt-funded AI tier — versus Microsoft 48, Google 56, Amazon 60, Meta 80, whose balance sheets barely move. The crack isn’t “AI”; it’s the leveraged periphery that financed compute with the ~$400B hyperscaler debt wave, meeting the highest global yields since 2008 (The era of free money is over — GlobalMktObserv’s line: for capital-intensive AI investment, debt is becoming expensive). Caveats: 162bp is still comfortably investment grade — “exploding” is zerohedge’s costume — and equity ignoring credit can persist for quarters. Watch the spread trend, not the level.
Update 08/2026 (Kobeissi): Oracle’s 5y CDS printed a record ~215bp (+70bp YTD), Meta 95, Nvidia 82, Broadcom +48bp — the repricing broadened from the leveraged periphery into the mega-caps as Big Tech issuance hit ~$200B YTD, nearly double all of 2025. The composition shift now shows in the IG issuer list itself (Big Tech becomes a borrower). kurtsaltrichter’s historical anchor for the pattern (https://x.com/kurtsaltrichter/status/2086804011080966345): bond spreads widened for months in 1873 before the panic ever hit the newspapers — “the S&P sits at a record while the bond market demands more to fund the same build. When those two disagree, credit is the one that tends to be right.”