CoreWeave's bonds slid to distressed levels in the week of October 12 to 16, and the pressure has moved past one company. On Thursday, October 15, CNBC reported that the cost of insuring CoreWeave's debt against default had widened to levels generally viewed as distressed. Yields on its 2031 bonds topped 15%.

This is a financing story, not a repeat of the equity selloff. AI data center debt is where lenders decide whether new projects get built at today's prices, and that market is now under strain. The IMF flagged a similar risk only days earlier.

How AI data center debt is getting harder to sell

On October 17, Bloomberg reported that junk-bond buyers were increasingly freezing out tech startups seeking financing. The report came after CoreWeave's co-founder said the company was exploring investment-grade borrowing. That would be a cheaper and more stable source of money than the high-yield market where its bonds now trade.

The shift matters because neoclouds like CoreWeave, which rent out GPU capacity to AI developers, have financed their expansion with debt. Their borrowing costs now shape how fast the next data centers get built.

How far CoreWeave's bonds and stock have fallen

The exact figures vary by outlet, and that matters for anyone tracking the market. One analysis, from Tech & Money Press, says CoreWeave's bonds, issued in June at a 9.675% yield, had risen to about 15% by mid-October. Startup Fortune puts the June coupon at 9.625% and says the 2032 bonds traded at a record 14.5%. Reports also differ on whether the distressed bonds mature in 2031 or 2032, so verify those details against CoreWeave's filings before you cite them.

The stock tells a similar story. According to Tech & Money Press, shares fell from about $91.70 on October 6 to a low of $59.51 on October 15, a drop of roughly 35% in eight sessions, before closing at $69.69 that day. BigGo Finance reports a different close, saying shares ended down 15.5% at $65.78 on Wednesday, October 14. It also says the Nasdaq Composite fell 1.4% intraday during that selloff. Check exchange data before you publish a specific closing price.

Debt is the other number to watch. CNBC describes more than $51 billion in debt and obligations used to finance CoreWeave's build-out. BigGo puts the figure at roughly $52 billion, including leases.

The $38 billion Oracle-linked loan lenders want to sell

JPMorgan and Mitsubishi UFG have spent more than six months trying to offload a record $38 billion construction loan tied to Oracle-leased data centers in Texas and Wisconsin, according to Startup Fortune. Some lenders are now selling slices at a discount. Only one outlet has reported these syndication details, and the date on that report is unclear, so treat the account as a lead that still needs a second source.

If the report holds, lenders are trying to reduce exposure they built during the boom. A Bank of America analysis, as reported by Yahoo Finance, flagged about $75 billion in AI data-center bonds and loans issued in September and October alone.

The pressure also shows up in deal pipelines. Reuters reported that listings for AirTrunk REIT, backed by Blackstone, Firmus, backed by Nvidia, and SB Energy, backed by SoftBank, were pulled or paused in the past week. Firmus had aimed to raise $5 billion while carrying about $30 billion of debt, according to The Next Web. The same report says Asian data-center M&A has reached $21.3 billion so far this year, compared with $10 billion in all of 2025.

What the IMF warned and why this is the test

The International Monetary Fund's October 13 Global Financial Stability Report named a correction in AI-linked stocks and credit as one of three risks that could sharply tighten financial conditions. The other two were rising sovereign yields and cyber threats, according to Unite.AI.

The IMF also examined how the money moves. It reviewed 14 publicly disclosed deals since 2023 in which hyperscalers led investors who put in about $300 billion of equity and $650 billion of debt into weaker AI firms. The deals came with $1.4 trillion of future revenue commitments, per Asia Times.

The IMF's term for the danger is wrong-way risk: investments lose value just as the revenue tied to them is impaired. Bond markets are now testing that logic in real time.

The IMF also framed this as a plausible scenario rather than a forecast of collapse, and it says banks remain broadly resilient in its stress test, according to MLQ. Hyperscalers now account for 10% of global investment-grade bond issuance, up from under 1% in 2024, according to figures cited in IMF coverage.

These warnings are not new. In January 2026, the Senate Banking Committee wrote to Treasury Secretary Bessent about AI debt and the risk that losses could spread through interconnected financial firms. USA 24 has also covered Fed Chair Kevin Warsh's warning that AI spending could shock the US economy.

Who feels the pain first: savers, projects and startups

Savers face the most direct exposure through funds. Reuters reported that one fund manager's firm holds CoreWeave bonds through an index-tracking ETF, which means investors who never bought an AI bond directly may still own the risk. John Han, a portfolio manager at F/m Investments, said: "We're definitely taking another look at our exposures across the AI infrastructure space."

Projects in Texas and Wisconsin could face tighter financing and possible delays, which would affect local construction jobs and grid plans. Startups that depend on junk bonds or private credit may find it harder to raise money. Businesses that rent AI computing from smaller neocloud providers may see harsher terms, and one analysis calls that group the most exposed.

Analyst Dan Ives estimated that 5% to 10% of planned projects will not get built, citing permitting and political pressure over power prices. That figure is an analyst opinion, not a measured number, but it shows how much some observers expect the build-out to shrink.

The earnings side matters too. AI Weekly reports that a Jefferies projection has consensus earnings growth for AI-related MSCI AC World stocks falling from nearly 100% in 2026 to about 30% in 2027. Lower growth leaves less room for error when companies carry heavy debt.

What to check in your funds and cloud contracts

This is general information, not investment advice. These checks take about ten minutes:

  • Open your 401(k), IRA or brokerage bond funds and read the top holdings for CoreWeave, Oracle or other AI-linked issuers.
  • Check whether any bond fund or leveraged ETF you own holds high-yield corporate debt, where the recent pullback is happening.
  • If your business rents AI compute from a smaller neocloud provider, ask about its financing, refinancing schedule and exit terms.
  • Ask how you would export your data, models and configurations if the provider's funding dries up, and get the answer in writing.

Do not assume a cloud contract is risk-free because the provider has a well-known name. A provider's balance sheet can change quickly once lenders step back.

What to watch next

Three developments will show whether this is a repricing or the start of a funding freeze. The first is whether CoreWeave can borrow at investment grade, as its co-founder said it was exploring. The second is whether the Oracle-linked loan clears and at what discount. The third is whether the stress reaches bank balance sheets or hyperscalers' own bonds. The IMF says banks remain broadly resilient, but that question is still open.

Two claims need confirmation. Jensen Huang reportedly said Nvidia does not need to rescue CoreWeave, but that account rests on a single source. Regulators are also looking at AI companies directly, as USA 24 has reported on the FTC's civil investigative demands to OpenAI and Anthropic.

The next few weeks will show whether lenders keep financing the build-out at current prices or start demanding more. For now, the bond market has put the AI buildout's money plan under its most public test so far.