According to Bloomberg, CoreWeave’s three co-founders have sold about $3.3 billion of stock since the March 2025 IPO—roughly $2.4 billion in 2026 alone. In nearly the same week, the firm’s five-year CDS spread briefly hit about 1,194.5 bps, implying a five-year default probability above 60%. The Nvidia-backed AI cloud is becoming a stress test for debt-fueled GPU buildouts.
What happened
– Selling pace: Michael Intrator has sold more than ~$907 million since Jan 1, 2026; Brannin McBee ~$1.06 billion since IPO (~$577 million this year, ~40.4% of his undiluted stake); Brian Venturo ~$734 million in Q2. – Company line: A spokesperson said the trio remain among the top six holders (~20% combined), sales were pre-arranged under 10b5-1 plans, and reflect personal financial planning. – Credit & shares: McBee talked up “infinite” insurer capital for expansion; CDS spiked the next day. Shares fell ~15.46% Wednesday and kept sliding; from the peak they are down more than 60%, wiping ~$15 billion of founder paper wealth while locking in ~$3.3 billion cash.
Debt is rising fast: combined debt and lease obligations ~$52 billion as of end-June; August CapEx guidance rose $4 billion to as much as ~$39 billion; Visible Alpha consensus sees ~$71 billion cumulative negative free cash flow for 2026–2028.
What “infinite capital” really means
McBee’s FT pitch centers on delayed-draw term loans collateralized by chip assets, plus power scale-up from ~1.5 GW toward 8+ GW by end-2030, funded partly by insurance balance sheets.
There is precedent: a March 2026 ~$8.5 billion loan backed by chips and a long-term Meta compute contract earned Moody’s A3 / DBRS A(low) / Fitch A-sf, anchored by Blackstone Credit & Insurance. The catch—ratings apply to a bankruptcy-remote SPV deal, not CoreWeave the parent. Corporate ratings remain speculative (Moody’s Ba3, S&P B+, Fitch BB-). McBee himself said loans backed by non-IG lab offtake are not yet insurance-eligible. “Infinite” capital is open mainly to IG-customer structures, not a blank check for the whole firm.
For related GPU-cloud economics on this site, see how profitable GPU rental really is and our earlier note on CoreWeave / Nebius “favorite sons”.
Value-destruction math and maturity mismatch
DA Davidson’s Gil Luria has called CoreWeave “subprime-like” for AI: ~9% borrowing cost versus ~1% returns—or even a 4–5% target that still fails to cover financing costs (“borrow at 9% to buy 5% Treasuries”).
CreditSights’ Jordan Chalfin flags long data-center leases versus short customer contracts: if capacity overbuilds, renewal leverage flips to buyers. Gartner’s October warning goes wider—over ~$3.5 trillion of AI infra spend by 2030 could face impairment if investment outruns revenue that can support it.
Huang’s optimism vs the CDS tape
Nvidia owns ~10% of CoreWeave, supplies GPUs, and has pledged ~$6.3 billion of cloud purchases—fueling “circular financing” worries. In Stockholm, Jensen Huang said CoreWeave “gets stronger every day” and that its cost of capital is falling, with no need for Nvidia bailouts—without hard numbers. Credit markets priced a different story via CDS.
FAQ
Q: Is the $3.3B sale illegal? A: Per the company, sales followed disclosed 10b5-1 plans as personal planning. The debate is scale and timing, not whether executives may sell at all.
Q: Does 60%+ CDS odds mean CoreWeave has defaulted? A: No. It is the swap market’s implied five-year default risk—a credit price, not a bankruptcy filing.
Q: Why does the Nvidia tie matter? A: Shareholder + supplier + large cloud buyer can inflate apparent demand if capital loops through “invest downstream → buy more chips.”
相关文章
