CryptoRevolution.Money

Macro

CoreWeave: The Needle in the AI Bubble?

By Tom Snellman ·

The debt market is pricing a better than 60% chance that CoreWeave, a key AI infrastructure company, defaults within five years. Most index funds are still full of the same AI names.

I have been tracking this for weeks. The pattern looks like a market that topped this summer and is now looking for the needle that pops the bubble. CoreWeave could be that needle. Here is how the pieces connect, what would confirm it, and what would prove me wrong.

Three pressure points in the AI boom

The question is not whether AI works. It is whether the money and the physics behind the buildout can keep up. Even Goldman Sachs, hardly an AI skeptic, points to three pressure points:

  • Power. Electricity, not chips, is now the main bottleneck. Goldman says grid capacity for 2026 and 2027 is essentially booked out, and the big cloud companies are already securing power for 2028 and beyond (investingLive). Goldman reads this as demand outrunning supply. I read it as a hard limit on how fast all that borrowed money can earn a return.
  • Debt. Big Tech is borrowing heavily to build data centers. Goldman Sachs Asset Management is underweight their bonds because it expects a wave of new issuance that could push up borrowing costs, even for the strongest names (Yahoo Finance).
  • Crowded trades. In August, leveraged money piled back into chip stocks. When they fell, automated selling made the drop worse. Goldman's trading desk called it a "liquidity trap" driven by money flows, not weaker fundamentals (BigGo Finance).

None of this means a crash is coming. It means the boom has little room for error. The rest of this post looks at where the cracks are already showing: first in Korea, then at CoreWeave.

The top came in the summer, and Korea saw it first

South Korea's Kospi peaked in June and is now down more than 40% from that high. I watch Korea because it is a chip market: Samsung and SK Hynix make up about half of the index, and they caused about 70% of the drop (Protos). These two companies build the memory that goes into AI servers. When demand for that hardware cools, Seoul tends to feel it before Wall Street does.

The first crack came on June 23, when the Kospi fell 10% in one day from a record high (Bloomberg). In October, foreign investors pulled more than $10 billion out of Korean stocks (Yahoo Finance).

US indices have held up better. The S&P 500 set its record close on August 13, and the Nasdaq briefly made a new high on September 21 (Reuters). That gap is the point: the chip market has already repriced, and the US market has not caught up yet.

CoreWeave: borrowing to rent out chips

CoreWeave is a "neocloud": a cloud company built only for AI. It buys Nvidia chips, puts them in data centers and rents the computing power to AI companies by the hour. The catch is timing. It pays for chips and buildings first and collects rent later, so it borrows to cover the gap.

The numbers from the second quarter of 2026 show how big that gap is:

ItemQ2 2026
Operating cash flow$679 million
Spending on equipmentabout $6.4 billion
Net interest expense$640 million, more than double a year earlier

Sources: KuCoin News, Startup Fortune.

Total debt reached $35 billion at the end of June, up from $21 billion at the start of the year. Interest alone now eats more than a quarter of revenue.

The stock is more than 60% below its June 2025 peak. The three co-founders have sold $3.3 billion of shares since the March 2025 IPO, $2.4 billion of it this year (Yahoo Finance). The company says the sales were scheduled in advance, and the founders still own about 20% (citybiz). Nothing illegal, but worth knowing.

What the debt market is saying

The bond market prices a 60% chance of default. Stock prices move on stories; debt prices move on one question: will I get my money back? That is why I watch two gauges.

Bond prices. A bond is a loan to a company that pays $1,000 back at maturity. When lenders doubt that payment, the bond trades below face value. CoreWeave's unsecured bonds due in 2032 fell below 80 cents on the dollar this week, the level widely seen as financial distress (Bloomberg). Buyers at that price earn close to 15% a year, up from about 12% two months ago.

Credit default swaps (CDS). A CDS is insurance against a company not paying its debt. The price is quoted in basis points per year: 100 bps means 1% of the insured amount. On October 15, five-year CoreWeave CDS hit a record 1,194.5 bps. That is almost 12% a year to insure the debt, and it implies a more than 60% chance of default within five years (Bloomberg).

One caveat: CDS-implied odds are "risk-neutral." They include the premium lenders charge for fear, so they tend to overstate the real-world chance of default (Wikipedia). Read them as a measure of fear, not a forecast. What matters most is the direction: a year ago the same gauge implied about 30% (Startup Fortune).

Why the Mag7 can't be isolated

The Magnificent 7 have strong balance sheets, but their AI story runs through the same money as CoreWeave's. Microsoft was about two-thirds of CoreWeave's 2025 revenue, and OpenAI holds up to about $22.4 billion in contracts with it (LongYield). Nvidia is CoreWeave's chip supplier, a customer and an investor all at once (MEXC News). That includes a $6.3 billion agreement to buy unused capacity through 2032 (Sacra).

The same dollars go around in a circle: Nvidia invests, CoreWeave borrows more against that, buys Nvidia chips, and rents them to Microsoft and OpenAI. If one link stops paying, the others feel it.

SegmentWhat drove the topHow stress moves onWhere it stands now
South Korean chips (Kospi, Samsung, SK Hynix)Memory demand for AI serversHardware demand cools first, so prices here turn firstDown more than 40% from the June peak
Neoclouds (CoreWeave)Debt-funded buildout of GPU data centersIf lenders stop lending, chip orders stopStock more than 60% below its peak; CDS imply above 60% default odds
NvidiaOrders from neoclouds and hyperscalersFewer orders and a supply glut weaken pricing powerSupplier, customer and investor in CoreWeave
Magnificent 7Huge AI spending priced as future profitIf the infrastructure layer can't pay its debt, that spending gets repricedUS indices still near their highs

The domino effect: from Seoul to Silicon Valley

Two dominoes have fallen. Nvidia and the Mag7 are next to watch.

  1. Seoul peaks: AI chip demand cools firstKospi down more than 40% since June; Samsung and SK Hynix drove about 70% of the fall.Happened: June 23 to now
  2. CoreWeave: the leverage trap closes$35 billion of debt, interest above a quarter of revenue; CDS imply above 60% default odds.Under way now
  3. Neocloud orders stop, and Nvidia feels itNvidia is CoreWeave's supplier, customer and investor; fewer orders weaken pricing power.Not yet: watch Nvidia's order commentary
  4. The Mag7 get repricedMicrosoft was two-thirds of CoreWeave's 2025 revenue; AI spending gets valued on cash flow, not hope.Not yet: US indices still near their highs

Solid cards: already happening. Dashed cards: not yet.

The chain runs in one direction: cooling chip demand, then lenders pulling back from the most indebted builders, then fewer orders for Nvidia, then a repricing of the companies that justified their valuations with AI spending. The first two are already in motion. The last two are what I'm watching.

What could prove me wrong

The bull case is real, and it deserves a fair hearing:

  • Contracted demand. CoreWeave reported a $104 billion backlog of signed work in August, roughly ten years of current revenue already under contract (LongYield).
  • Strong backers. Nvidia CEO Jensen Huang said this week that CoreWeave's position is stable and improving. His comments helped the stock recover on Thursday (CNBC).
  • Fear is not default. CDS odds measure fear. A company can trade at distressed levels and still refinance.

What would change my view: CoreWeave raising new debt at lower rates, its bonds recovering above 90 cents, or the Kospi reclaiming its June high.

What I'm watching, and what it means for Bitcoin

Three gauges tell me whether the next domino is falling:

  1. CoreWeave's 2032 bonds. Below 80 cents means stress is still building. A move back above 90 would ease it.
  2. Five-year CoreWeave CDS. Still rising means lenders are getting more nervous, not less.
  3. Kospi and the chip stocks. Korea led on the way down. It will likely lead on the way back up too.

Bitcoin trades as a risk asset, so an AI repricing would hit crypto as well. My BTC cycle model points to the next accumulation phase starting around November 18, 2026. But the bottom could just as well stretch into Q1 2027. Nobody knows the exact turning point in advance, and anyone who claims to is guessing. That is why I follow the market every week instead of betting on one date.

This post is for information only and is not financial advice.

Sources