Is CoreWeave the New Bear Stearns?

The AI cloud darling just dropped 15% in a day, and its bonds now trade like junk. We read the filings to see if the 2008 comparison holds.

March 12, 2008. Bear Stearns CEO Alan Schwartz goes on CNBC and says the firm sees no pressure on its liquidity. Four days later, Bear sells itself to JPMorgan for $2 a share. A year earlier the stock had been at $172.

October 14, 2026. CoreWeave CEO Mike Intrator writes to staff that wider market fears “are bleeding into the narrative around CoreWeave,” and says investors “misunderstand and act irrationally.” That same Wednesday, CRWV closes down 15.5% at $65.78 on 107 million shares, more than triple its normal volume. Its unsecured 2032 notes trade below 80 cents on the dollar. That’s a yield near 15%.

So is CoreWeave the new Bear Stearns? Partly. The balance sheet rhymes, loudly. The way it’s funded doesn’t. And that difference decides everything.

The Tape | What Just Happened

CoreWeave IPO’d at $40 in March 2025. It ran to $187 on June 20, 2025. It now trades around $65, roughly 65% off the top and down 8.1% year to date, while the QQQ is up 18.9%.

This week’s break had several triggers stacked together. A venture capitalist flagged the company’s bonds on X late on October 13. Gartner analyst Will Sommer said more than 80% of neocloud providers may not survive as independent businesses much past 2030 if AI spending keeps its current pace. And CoreWeave set up an at-the-market program to sell up to 35 million new shares, less than a month after pricing $3.7 billion of 2.875% convertible notes due 2033.

That convert is worth a closer look. Its conversion price is $97.85, set when the stock was $79.88. Bondholders who bought in September are already well out of the money on the equity kicker.

When the stock and the bonds start telling the same story, listen.

Partner | Trade the Blowup Both Ways

A quick pause before we get into the balance sheet, because this one’s live.

CRWV just dropped 15% in a session, and its bonds are trading like a distressed credit. Whatever you take from the rest of this piece, you’ll want to express it the day you finish reading. That might be a long on the backlog, or a short on the debt stack.

That’s what Liquid is for.

Liquid is a non-custodial trading app that runs perpetual markets on things that normally need five different accounts: US and international stocks, indices, commodities, FX, crypto, prediction markets, and pre-IPO names like OpenAI and Anthropic. It’s one interface, and you can go long or short. You fund it by card, bank or wallet, and your keys stay yours. Crypto perps trade 24/7. Everything else follows its underlying venue’s hours.

For this story, CoreWeave trades on Liquid as CRWV, right next to its supplier, shareholder and customer NVDA. If you think the $104B backlog bails out the balance sheet, that’s a long. If you think the 2032 bondholders already called it, that’s a short. Liquid advertises up to 50x, but the maximum multiplier varies by market, so check the live ticket on CRWV before you size anything.

70,000+ traders. $5B+ in trades.

Get set up here: liquid.trade/r/VHLA

Paid partnership with Liquid.

‍

Rhyme #1 | Leverage on Collateral That Ages

Bear ran at 34.6x leverage: about $350 billion of assets on $11.1 billion of equity at the end of 2006. Much of that was mortgage paper whose value was a matter of opinion, until the day it wasn’t.

CoreWeave carried $35.1 billion of debt on its balance sheet at June 30, up from $21.4 billion six months earlier. That’s 64% more debt in half a year. Add $16.3 billion of operating lease liabilities and you’re at about $51.4 billion. The company’s whole market cap at Wednesday’s close was about $35.9 billion. The debt alone roughly matches what the equity is worth.

The collateral is $46.7 billion of property and equipment, mostly GPUs and data center build-out, depreciated over the six-year useful life CoreWeave uses for its servers. Meanwhile Nvidia ships a new architecture about once a year. The bull case needs a chip bought in 2026 to keep earning into 2032.

Now the Q2 income statement. Revenue was $2.58 billion, up 112%. Adjusted EBITDA was $1.51 billion, a 59% margin. Then depreciation shows up. GAAP operating income was a $49 million loss. Net interest expense was $640 million. The net loss was $626 million. Interest ate five times the $128 million of adjusted operating income.

Adjusted EBITDA is the brochure. Interest expense is the bill.

Rhyme #2 | The Canary Is Coughing

Bear’s warning came nine months early. In June 2007, two Bear-sponsored hedge funds, levered into subprime structured credit, blew up. Investors lost about $1.8 billion, and the funds filed for bankruptcy on July 31, 2007. The Street called it contained.

CoreWeave’s canary is what its money costs. Look at its delayed-draw term loans, which are GPU-backed and sized against customer contracts:

DDTL 4.0 (March 2026): SOFR + 2.25%

DDTL 5.0 (May 2026): SOFR + 4.50%

DDTL 5.5 (August 2026): SOFR + 5.50%

The structures differ (4.0 has only limited recourse to the parent), but every step costs more. The $2.6 billion loan closed in August needed its spread widened by 100 to 125 basis points to get done. The oldest facility, DDTL 1.0, carries an effective rate of about 15%. Credit default swaps traded around 855 basis points back in August.

And the backdrop isn’t helping. The 10-year Treasury sits at 5.19%. September CPI was 3.7% year over year. Leveraged buildouts thrive when money is cheap, and right now it isn’t.

The bond desk reads the 10-Q before the stock desk does.

Rhyme #3 | Three Customers, 72% of Revenue

According to the Q2 10-Q, Customer A was 36% of revenue, Customer B 26%, and Customer C 10%. That’s 72% from three names. To be fair, this is progress: Microsoft alone was 67% of revenue in 2025. Two customers each account for 32% of receivables.

The named heavyweights are Microsoft, Meta and OpenAI. Meta has signed roughly $21 billion running to December 2032, on top of an earlier deal of up to $14.2 billion. OpenAI’s contracts total up to about $22.4 billion. OpenAI pays its compute bills from money it raises, so that’s counterparty risk stacked on counterparty risk.

Bear’s counterparties could walk out the door overnight. CoreWeave’s are signed for years. That’s a real advantage. It’s also only as good as the balance sheets behind the signatures.

The backlog is only as strong as the weakest signature on it.

Where It Breaks | No Overnight Repo, No Four-Day Death

Here is where the comparison falls apart.

Bear funded a long-dated balance sheet with overnight money. By October 2007, State Street would only lend to it overnight. In March 2008, Bear’s liquidity pool went from about $18 billion on March 10 to about $2 billion by the night of March 14. Lenders simply stopped rolling over their repo. It took four days.

CoreWeave’s debt is term debt. The delayed-draw loans mature between 2030 and 2032 and sit in separate subsidiaries. Each one is secured by specific GPUs and specific customer contracts, mostly with investment-grade counterparties. The senior notes run from 2030 to 2032, and the converts to 2031 and 2033. The revolver has $2.0 billion available and runs to November 2029. There’s $5.5 billion of cash. No lender can pull CoreWeave’s funding on a Thursday afternoon.

Then there’s the backlog: $104 billion at June 30, plus more than $25 billion of new commitments early in Q3. About 41% of it, roughly $42.5 billion, is scheduled to turn into revenue within 24 months.

You can’t run on a GPU the way you run on a repo line.

The Real Risk | The Capital Treadmill

If CoreWeave breaks, it’ll break slowly. The risk is a squeeze, not a stampede.

CoreWeave guided 2026 capex to $35 billion to $39 billion, against revenue guidance of $12.4 billion to $13.2 billion. That’s about 3x revenue going out the door. Trailing free cash flow was negative $13.7 billion. In Q2 alone it borrowed $13.5 billion and repaid $3.9 billion. Scheduled principal payments are $4.4 billion for the rest of 2026 and $6.2 billion in 2027. Current liabilities of $20.9 billion sit against $9.5 billion of current assets.

Every dollar of growth needs a new dollar of capital, and that dollar keeps getting more expensive. If the window narrows, management has levers: slow the build, sell shares through the ATM (35 million shares at Wednesday’s close is about $2.3 billion), or issue more converts. Every one of those levers costs shareholders.

Bear’s equity got wiped out because the firm ran out of cash over a weekend. If CoreWeave gets squeezed, the more likely path is dilution, delayed buildouts and a long grind.

Bear was a heart attack. CoreWeave’s risk is blood pressure.

The Checklist | What Decides It

CoreWeave’s Q3 report is expected around November 11 (the date isn’t confirmed yet). Here’s what matters:

Revenue against guidance. The Q3 guide is $3.4 billion to $3.6 billion. A miss here would hit harder than any other.

The interest run rate. Q2 annualizes to about $2.56 billion. Watch which way it moves.

Capex. Any trim to the $35 billion to $39 billion guide tells you how management reads its funding window.

The next loan’s spread. If SOFR + 5.50% becomes the floor, the treadmill is getting steeper.

Concentration. Does the top three keep falling from 72%?

The 2032 bonds. Back above 80 cents, or lower?

ATM usage. How many of the 35 million shares actually got sold.

Verdict | It Rhymes, It Doesn’t Repeat

Is CoreWeave the new Bear Stearns? On leverage, on collateral that loses value fast, and on a CEO calling the market irrational while his bonds disagree, the rhyme is loud.

On the thing that actually killed Bear, overnight funding that disappeared in four days, the answer is no. CoreWeave’s lenders are locked in for years, and so are its customers.

That buys time. But time only helps if the economics work before the next refinancing bill arrives. If the GPUs pay back their cost before they’re obsolete, this goes down as an aggressive build that worked. If they don’t, the 2032 bondholders have already told you how this movie ends.

Bear ran out of time. CoreWeave bought some. The question is whether the GPUs pay for it first.

Not financial or investment advice. Do your own research.

‍