CoreWeave bonds vs AI IPOs

by Huda Thomas: CoreWeave

Firmus had an Nvidia-backed IPO and walked away from it because buyers wouldn't pay the ask. That's the moment I keep circling back to, since it sits right beside the debt side of the AI trade telling a completely different story. CoreWeave leases compute to Meta, OpenAI and Microsoft. The cost of protecting its debt against default climbed to a record, one that implies better than a 60% chance of default inside five years. Bonds maturing in 2031 slipped to roughly 82 cents on the dollar, and the yield landed near 15%. Two months earlier that same yield sat closer to 12%. Allianz's chief investment officer told the Financial Times that data-center bonds have gotten to the point where only the strongest borrowers make the cut. Goldman Sachs says hedge funds unloaded AI stocks at the second-quickest clip in more than a year. Then look at what's being sold to the public. SpaceX pulled in roughly $86 billion from its IPO in June. Anthropic is prepping a November listing that could bring in as much as $100 billion at a $2 trillion valuation. OpenAI wants $30 billion at a $1.4 trillion valuation. And per the IMF, single stocks now account for 28% of the geared ETFs retail traders lean on to magnify bets, up from 16%. So the lenders, the ones who only collect if the cash actually comes back, are backing off, while the biggest share sales ever assembled get pointed at pension funds, index funds and regular traders. I'd side with the lenders in this one, and here's the logic. A stock buyer can sit around for years waiting on a story to pan out. A lender needs the money to arrive on a schedule, so checking whether it will show up is basically the job. When lenders start asking for 15%, they're telling you what they found. Rewind to June 2007. Bear Stearns put up as much as $3.2 billion to keep one of its subprime hedge funds afloat, and by mid-July both funds had shed nearly all their value. Credit traders caught the damage that summer while stock investors kept buying. The S&P 500 closed at a record 1,565.15 on October 9, 2007. By March 2009 it sat at 676.53. The bull case isn't nothing. CoreWeave carries a $104 billion backlog, its top line grew by more than double in the prior quarter, and Nvidia has committed to buying any computing capacity it can't rent out through 2032. Anthropic reports an adjusted profit. Goldman reads the selling as money rotating out of smaller AI names and into the giants. Except the giants have their own issue. By Bank of America's consensus estimates, the five biggest AI spenders swing from $191 billion in combined free cash flow last year to negative $63 billion this year. Rotating into companies that are about to burn cash on AI still hinges on someone lending them the gap. So keep an eye on the bonds first. If CoreWeave's debt keeps sliding and more data-center borrowers get pushed into double-digit yields, the November IPOs will be selling AI to the public right at the top. If the bonds bounce back, I'll say the lenders got spooked early. In 2007 the warning surfaced in the bond market three months before stocks peaked. Ignoring it a second time would be a choice. fr