The Credit Repricing: CoreWeave's Bonds Go Distressed and the AI Selloff Escapes the Neoclouds
The week of October 13–17, 2026 was the AI capex cycle's financing-layer stress test: the selloff stopped being about chip demand and became about who funds the buildout, and at what cost. The trigger was CoreWeave — about $50 billion in debt, April-issued bonds trading around a 15% yield (last 79.15 cents on the dollar per Tradeweb), and default insurance at ~1,000 basis points, implying a ~63% chance of default within five years, up from about 30% a year ago — despite a Nvidia agreement to buy any unsold CoreWeave capacity through April 13, 2032 (Reuters, Oct 15). CoreWeave raised its annual capex forecast by $4B to as much as $39B in August; analysts expect ~$13B revenue and a $2.13B loss this year.
The rout then spread to names with no financing problem. Per Forbes (Jim Osman, Oct 18): Dell lost ~23% in the week; CrowdStrike, Palo Alto Networks and Snowflake fell 17–18%; on Friday Micron fell 5.6%, AMD 6%, Intel 5.8%, Broadcom 3.6%. The Nasdaq finished the week down 4.5% and the Philadelphia Semiconductor Index suffered "its longest losing streak since 2008" — Reuters counts $1.6 trillion of SOX market value erased in an 11.6% slide over six sessions. Nvidia itself closed at $214.13 (−6.6% over five days, −10.5% off its 52-week high, $5.17T market cap) (NVDA market view).
The bank layer is the new evidence. JPMorgan, Morgan Stanley, SMBC and MUFG are exploring significant risk transfers, and one group including JPMorgan and MUFG has spent more than six months trying to distribute $38 billion of construction debt tied to a Texas and Wisconsin data center project leased to Oracle, with some lenders willing to sell pieces at a discount to non-bank buyers (Startup Fortune, on FT/Bloomberg reporting). Matthew Moniot, co-head of credit risk sharing at Man Group, on the deal sizes: "out of scale to anything we've thought about, ever." Oracle's five-year CDS climbed to ~151bp in early October — levels last seen in 2009 — after S&P cut it to BBB− in July; Oracle is targeting more than $40B of additional debt and equity this fiscal year. Hyperscaler bond order books shrank from ~5x covered in February to under 2x by July, and CoreWeave-leased paper prices at 7.00–9.875% vs 5.70–7.875% for hyperscaler-leased bonds (CCIR data via the same report) — capital is rotating to IG-hyperscaler-leased data centers and away from weaker credits.
The issuance channel is closing too: AI-related debt issuance fell to $23B in September — less than half of August, and down from June's record $113B (Morgan Stanley data), with US investment-grade AI-related bond issuance stopping completely in September (HN discussion of the FT report, 385 points). As one commenter put it, "syndications usually clear in a few weeks but this has been stuck in distribution for six months and still not closed" (HN comment). The 10-year Treasury hit 5.304% this month — its highest since 2002 — after a $39B auction cleared at 5.30%, the highest auction yield since 2000; the Fed's September minutes named AI-related corporate borrowing as a contributor. The macro backdrop now: 10-yr at 5.1% (Oct 15), fed funds 3.75%, with the FOMC due Oct 28 (macro view).
Gartner, published the same day as CoreWeave's trough: ~80% of neoclouds could shut down by 2030 and more than $3.5T of AI spending could face impairment. Analyst Will Sommer: "In 2026, AI providers are expected to spend nearly $1 trillion on AI infrastructure, and are forecast to invest an additional $6 trillion between 2027 and 2030... our research indicates enterprise and consumer AI revenue will be insufficient to support profitable returns on all planned AI capital expenditures during this period" (Reuters).
The demand side pushed back. Jensen Huang — Nvidia holds ~10% of CoreWeave — said "CoreWeave's financial strength is improving every day," adding "I'm optimistic. Their cost of capital is actually decreasing," and that he sees no need for Nvidia to backstop the company. IG Group's Axel Rudolph located the divergence precisely: "It looks more like investors questioning how the build-out is being funded than a sudden collapse in appetite for AI computing. But if the cost of capital keeps rising, even strong demand may not be enough to protect the most heavily leveraged players."
Meanwhile the physical layer reported sold-out and got punished anyway: TSMC TSMC Q3 2026: Sixth Straight Record Quarter, 2nm at 8% of Wafers — and the Stock Falls Anyway and ASML ASML Is Now the Bottleneck: Supply Falls Short Through 2028, Orders Discussed for 2029–2030 delivered record beats the same week, and both stocks fell. Forbes: "TSMC and ASML delivered the kind of numbers that should have lifted their shares. The response was weak. When a stock stops rewarding an earnings beat, investors are usually worried about something that has not reached the reported numbers yet." The project layer is cracking too: the FT counts at least 47 major data-center projects delayed or cancelled in 2026 (analysts regard only ~35% of the proposed ~$10T pipeline as credible), and Microsoft reportedly withdrew as anchor tenant from an Nscale project in Essex because grid power might not arrive "until sometime in the next decade" (Forbes).
What it means: the market is repricing the buyers of chips, not the chips. Demand signals (sold-out foundry and EUV capacity, record beats) and financing signals (distressed neocloud credit, bank risk transfers, collapsed issuance) diverged sharply this week. The near-term tests: Meta reports Oct 21, Microsoft/Alphabet Oct 28, Amazon Oct 29 — capex revisions there either validate the ~$725–750B 2026 spend or mark the top (Hyperscaler Capex: Five-Way 2026 Total Now ~$800B; Cloud Revenue Growth Is the Counterweight to Funding-Layer Stress). This extends OpenAI's Revenue Run-Rate Gap: ~$50B vs the ~$70B Circulated — and What It Did to the Capex Chain (the OpenAI revenue-gap trigger) and Oracle: Q1 FY2027 Capex ($28.5B) Exceeded Operating Cash Flow ($23.1B) — and the Stock Has Now Lost 54% From Its High (Oracle's funding math) into the credit market. The Anthropic IPO, pending after Q3 revenue rose 45% to nearly $17B, is the next window test (Reuters).