CoreWeave's June bond has gone from yielding 9.625% to 15%

Treasury yields did little on Wednesday. The 10-year rose 3.6 basis points and the two-year 7.2 ^[1], and the 30-year touched 5.63% before pulling back sharply in the afternoon ^[2]. A CoreWeave bond issued in June at a 9.625% yield recently yielded 15% ^[3], a gap of 5.375 percentage points, or 537.5 basis points ^[31]. The one-day Treasury moves are small beside that gap, but the gap has built since June, and Treasuries may have contributed over the longer span.

Rates did pick the day, because the prospect of a second Fed hike as soon as this month rose after hot consumer price data ^[4]. Whether rates also set the price is a separate question. We think the split between "rates" and "weak demand" is the wrong one for a borrower that has issued $22 billion of bonds and $26 billion of loans, according to LSEG data ^[5], or $48 billion in all ^[32]. The premium over rates is the lender's estimate of whether revenue will cover the build (or, more precisely, the interest on the build), and that estimate is itself a judgement about demand.

What the 15% is made of

CoreWeave's $1.25 billion of unsecured bonds due 2032 touched 15% intraday and closed near 14.5% ^[6]. Set that close against the 30-year's 5.63% peak and the gap is 8.87 points ^[33]. The maturities differ, so the figure gives an order of magnitude only. Still, for the Fed to explain most of a 14.5% yield, Treasury yields at the relevant maturity would have to sit far above anything the reports cite.

Every $1 billion raised at 15% instead of 9.625% adds about $54 million a year in interest ^[34]. Crypto Briefing wrote that higher yields on existing bonds tend to bleed into the cost of any new debt, and that a company that needs to keep borrowing to keep building feels that directly ^[7].

Credit moved first

Crypto Briefing set out the sequence as credit markets moving, then stocks, then hedging demand ^[8]. CNBC agrees on the first step: neocloud shares plunged after CoreWeave's debt yields continued to reach new heights ^[9]. Default insurance moved too. The cost of insuring debt against default for some AI companies, including Oracle and CoreWeave, surged to record highs, according to S&P Global Market Intelligence data ^[11].

Oracle is the useful comparison. The early CNA report called it debt-laden and said most of its dollar bonds had declined ^[12]; its shares were down 9% in the later report ^[15]. Alphabet and the other major hyperscalers were mostly down around 2% ^[13].

The single-stock percentages differ by report and by hour. An early CNA report had CoreWeave down 9.5% and Nebius down 15.1% ^[14]. Its later report had CoreWeave down 16% and Nebius down 11%, with some $80 billion of market value erased across the leading decliners it named ^[15]. The two names swap places, so the $80 billion aggregate is the sturdier number.

The demand evidence is about returns

Several voices on the day questioned returns, which is a narrower claim than weak demand. Kim Forrest, chief investment officer at Bokeh Capital Partners, said: "We're getting a whole lot of information suggesting that the largest AI companies may not be generating the earnings or revenue people anticipated, and I think that makes everyone nervous about companies like CoreWeave." ^[16] Barclays downgraded Dell and two other enterprise hardware stocks, noting that Dell shares have quadrupled this year ^[18], and said: "Valuations have run, commodity costs have helped revenues, and we don't see a meaningful AI inflection yet." ^[17]

Gartner said up to 80% of neocloud providers might be unable to keep operating as independent companies by 2030 if their operating revenue fails to match their planned capital expenditures ^[19]. In our view that is also a financing test: planned capital expenditure depends on what the money costs, so a neocloud with intact demand can still miss it at 15%. CoreWeave chief executive Mike Intrator said in an internal memo that the market is acting "irrationally" and that he remains confident in the company, Bloomberg reported ^[20].

Neither quote, as reported, cites an order book, contract or utilisation figure for a neocloud, and the reports include no June Treasury yields, no coupons or maturities for CoreWeave's older debt, no neocloud backlog and nothing on the state of the equity listing market. That leaves the split between rates and demand to be argued from timing and from the size of the premium.

Who can fund the next build

The window was narrowing before the inflation report. AI-related corporate debt issuance fell to $23 billion in September from a $113 billion peak in June ^[21], a drop of 79.6% in three months ^[35].

Crypto Briefing wrote that investors seem to be drawing a line between companies that fund expansion from cash flow and those that depend on capital markets ^[22]. Jed Ellerbroek, a portfolio manager at Argent Capital Management, sized the first group. Meta, Microsoft, Google, Amazon and SpaceX, he said, "are spending like 75 per cent of the money that's being invested in data centers here in the country" ^[23]. By his count that leaves roughly a quarter for everyone else ^[36].

That quarter is where Ross Mayfield, investment strategy analyst at Baird, places the strain. He told Reuters that long rates "above 5 per cent and creeping towards 6 per cent" are "a problem for the marginal borrower." He added: "It gets a lot harder for the tiers below the kind of big four hyperscalers." ^[24] Ellerbroek described the lenders on the other side this way: "Investors have definitely not lost their minds. They're being very cautious and I think selective." ^[25]

The Financial Times reported, according to CNBC, that CoreWeave is targeting insurance companies for further debt financing ^[10]. CNBC said the article apparently did not help the shares.

Three ways this is wrong

Rates may account for more than we allow, since the 30-year reached 5.63% and Mayfield puts the strain at long rates above 5% ^[2]. If CoreWeave's bonds retrace as long yields settle, the premium was mostly the curve. Demand may be weaker than the reports can show. If issuance stays near $23 billion once yields settle, lenders are pricing returns, and the Gartner warning and Forrest's remark become the base case.

Flow may be doing more of the damage than either. Options volume in CoreWeave ran more than seven times its 30-day average, and traders bought just shy of 200,000 puts against 106,000 calls ^[26], nearly two puts for every call ^[37]. The largest single trade CNBC cited was 26,000 puts at a 37.5 strike expiring Dec. 18, opened near the ask at 74 cents apiece for $1.9 million, needing a further 42% fall to pay off ^[27]. The strike sits 6.25% below the $40 IPO price ^[38][28]. Scott Bauer, CEO of Prosper Trading Academy, called the trade "likely some kind of margin play for someone to hedge their exposure in highly correlated stocks or maybe even the Nasdaq." ^[29] A hedge of that kind shows someone wanted protection and tells us little about contracts.

Chamath Palihapitiya posted questions about a private credit bubble alongside a screengrab of CoreWeave's bonds at a 15% yield, and Elon Musk replied "yikes." ^[30]

Our view is that rates picked Wednesday, that the premium is CoreWeave's own credit, and that weak demand is still untested. If the 2032 bonds go back toward the 9.625% at which the June issue priced while long yields stay near their peak, we over-read credit. If they stay near 14% after the long end retraces, the credit reading holds and the rates story loses.